A method, system and storage medium for managing fever in children throughout the entire course of the disease

Abstract

The application provides a child fever whole course management method, system and storage medium, the method comprises the following steps: acquiring multiple rounds of conversation information related to child fever, identifying a danger signal based on the multiple rounds of conversation information and generating a triage result containing a medical history abstract and a triage level; synchronizing the triage result to a doctor terminal when seeing a doctor, and acquiring a diagnosis result input by the doctor based on the triage result; generating a corresponding post-diagnosis follow-up plan based on the diagnosis result and the triage level; in the post-diagnosis stage, acquiring follow-up data reported based on the post-diagnosis follow-up plan, calculating a disease deviation degree according to the follow-up data, the medical history abstract and the diagnosis result; adjusting the post-diagnosis follow-up plan and issuing a warning when the disease deviation degree reaches a preset trigger condition; and generating a health record in units of fever events based on the multiple rounds of conversation information, the triage result, the diagnosis result and the follow-up data. The application can improve the continuity and individualization level of the disease course management.

Description

Technical Field

[0001] This application relates to the field of disease management technology, and in particular to a method, system and storage medium for the whole course of fever management in children. Background Technology

[0002] Fever in children is one of the most common reasons for seeking medical attention in pediatrics. Currently, hospital information systems generally adopt a segmented management model, where registration and triage, doctor consultation, and discharge guidance operate independently. With the application of artificial intelligence in the medical field, some solutions have attempted to introduce AI into pediatric diagnosis and treatment, such as conducting initial consultations through natural language interaction or providing auxiliary decision-making references during diagnosis. However, existing systems are essentially still built around a single medical visit event, with data generated at each stage stored in different modules, lacking a unified disease management mechanism centered on the individual child and organized longitudinally along the time dimension. This segmented and fragmented management model makes it impossible to form a continuous and traceable disease record for the entire process from before the visit to after discharge, and also makes subsequent follow-up plans dependent on manual formulation and unable to be dynamically adjusted as the condition evolves, making it difficult to achieve closed-loop management of the entire disease course. Summary of the Invention

[0003] To achieve standardized, dynamic, and closed-loop management of children with fever throughout their entire course of illness, from pre-visit medical attention to recovery, and to improve the continuity and personalization of disease management, this application provides a method, system, and storage medium for the full-course management of childhood fever.

[0004] Firstly, a method for managing fever throughout the entire course of childhood illness is provided, the method including: Acquire multi-turn dialogue information related to fever in children, identify danger signals based on the multi-turn dialogue information, and generate triage results including medical history summary and triage level; During a medical visit, the triage results are synchronized to the doctor's end, and the doctor's diagnostic results based on the triage results are obtained; A corresponding post-diagnosis follow-up plan is generated based on the diagnostic results and the triage level; In the post-diagnosis stage, follow-up data reported based on the post-diagnosis follow-up plan is obtained, and the deviation of the condition is calculated based on the follow-up data, the medical history summary, and the diagnosis results. When the deviation of the condition reaches the preset trigger condition, the post-diagnosis follow-up plan is adjusted and an early warning is issued. Based on the multi-round dialogue information, the triage results, the diagnosis results, and the follow-up data, a health record is generated based on the fever event.

[0005] In some embodiments, identifying danger signals based on the multi-turn dialogue information and generating a triage result including a medical history summary and triage level includes: The system acquires the response text and fever-related time-series data from the multi-turn dialogue information, extracts positive symptom entities from the response text layer by layer according to the preset consultation skeleton, and generates a body temperature change trend based on the fever-related time-series data. Based on the positive symptom entities and the stated body temperature change trend, update the danger signal flag bit vector according to the danger signal list, and record the first occurrence time and confidence level of each danger signal; Based on the positive symptom entities, the body temperature change trend, the danger signal marker vector, and the confidence level, a base score, a danger signal weighted value, and an uncertainty penalty value are calculated. The base score, the danger signal weighted value, and the uncertainty penalty value are then superimposed to obtain an urgency score. Based on the urgency score, the corresponding triage level is matched, and the positive symptom entities and body temperature change trends extracted in each round are combined in time series to form a medical history summary.

[0006] In some embodiments, generating a corresponding post-diagnosis follow-up plan based on the diagnostic result and the triage level includes: Obtain the disease type and severity grade from the diagnostic results, as well as the child's age risk level represented by the triage level. Input the disease type, the severity grade, and the child's age risk level into a preset pediatric fever course state machine template to instantiate and generate an individualized state machine that includes the acute fever phase, fever reduction observation period, recovery period, symptom recurrence pending judgment period, deterioration warning period, and event closure period, as well as their state transition conditions. Based on the preset duration and preset follow-up touchpoints of each state in the individualized state machine, a nursing content identifier and a self-assessment questionnaire identifier corresponding to each preset follow-up touchpoint are generated, and a post-diagnosis follow-up plan is formed according to the time sequence of each preset follow-up touchpoint and the corresponding nursing content identifier and the self-assessment questionnaire identifier.

[0007] In some embodiments, calculating the deviation of the condition based on the follow-up data, the medical history summary, and the diagnostic results includes: Extract body temperature time-series data, symptom quantitative scores, and new symptom entities from the follow-up data; extract historical body temperature peaks from the medical history summary; and match the corresponding expected recovery curve and expected symptom improvement trajectory for the same disease from the diagnostic results. Calculate the dynamic time curvature distance between the body temperature time series data and the expected recovery curve of the same disease, and normalize the dynamic time curvature distance to obtain the time series deviation. Calculate the absolute value of the residual between the symptom quantitative score and the expected symptom improvement trajectory at the same time point, and then normalize the weighted sum of the absolute values ​​of the residuals at each time point to obtain the symptom outcome deviation. The newly emerging symptom entity is matched with a pre-set list of danger signals. When the match is successful, the penalty value of the newly emerging danger signal is calculated according to the preset weight of the matched danger signal. When the match fails, the penalty value of the newly emerging danger signal is set to zero. The disease deviation is obtained by weighted summing of the time sequence deviation, the symptom outcome deviation, and the penalty value for newly emerging danger signals.

[0008] In some embodiments, adjusting the post-diagnosis follow-up plan when the deviation of the condition reaches a preset trigger condition includes: The deviation level is obtained by comparing the degree of deviation of the condition with a preset multi-level deviation threshold. When the deviation level is mild, the post-diagnosis follow-up plan remains unchanged; When the deviation level is moderate, a temporary follow-up contact is inserted between two adjacent preset follow-up contacts in the post-diagnosis follow-up plan, and close observation guidance is pushed out at the temporary follow-up contact. When the deviation level is significant, the individualized state machine is forcibly migrated to the deterioration warning period. Based on the deterioration warning period, the preset follow-up touchpoints that are not currently executed in the post-diagnosis follow-up plan are suspended, and a high-frequency monitoring scheme is activated.

[0009] In some embodiments, the method further includes: After the high-frequency monitoring scheme is activated, the medical history summary, diagnosis results and reported follow-up data corresponding to the current fever event are retrieved, and the disease retrospective summary is generated by merging them according to the preset retrospective summary template. The disease retrospective summary is then pushed to the bound doctor's terminal.

[0010] In some embodiments, generating health records based on the multi-turn dialogue information, the triage results, the diagnostic results, and follow-up data, using febrile events as units, includes: The multi-turn dialogue information, the triage results, the diagnosis results, and the follow-up data are written into the event log indexed by the fever event ID in chronological order to obtain the event log stream; The event log stream is aggregated by time dimension into a materialized view with the child ID as the primary key to obtain a health record. Each fever event record in the health record includes a symptom evolution map corresponding to the multi-turn dialogue information, an emergency change log corresponding to the triage results, a diagnosis conclusion corresponding to the diagnosis results, and a body temperature curve and outcome status corresponding to the follow-up data.

[0011] In some embodiments, the method further includes: Obtain the pre-diagnosis triage direction and the final diagnosis during diagnosis for the same fever event in the health record, perform structured alignment of the pre-diagnosis triage direction and the final diagnosis during diagnosis, and extract the difference features to obtain the diagnostic deviation set; Based on the diagnostic deviation set, a correction feedback package containing field-level modified records and diagnostic direction deviations is generated. The correction feedback package is then sent to the pre-diagnosis engine to update the pruning threshold and follow-up question trigger weight of the decision tree node.

[0012] Secondly, a comprehensive management system for the entire course of fever in children is provided, the system including: The identification module is used to acquire multi-turn dialogue information related to fever in children, identify danger signals based on the multi-turn dialogue information, and generate triage results including medical history summary and triage level; The consultation module is used to synchronize the triage results to the doctor's end during the consultation and to obtain the diagnosis results input by the doctor based on the triage results; The follow-up module is used to generate a corresponding post-diagnosis follow-up plan based on the diagnosis results and the triage level. In the post-diagnosis stage, it obtains follow-up data reported based on the post-diagnosis follow-up plan and calculates the degree of deviation of the condition based on the follow-up data, the medical history summary and the diagnosis results. The early warning module is used to adjust the post-diagnosis follow-up plan and issue an early warning when the deviation of the condition reaches a preset trigger condition; The management module is used to generate health records based on the multi-round dialogue information, the triage results, the diagnosis results, and the follow-up data, with fever events as the unit.

[0013] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon that can run on a processor, wherein when the computer program is executed by the processor, it implements a method for managing the entire course of fever in children as described in the first aspect.

[0014] The present invention has at least the following beneficial effects: Firstly, by organically linking pre-diagnosis intelligent triage, in-diagnosis data synchronization, and post-diagnosis adaptive follow-up, and taking fever events as the main thread to connect all links, the problem of data fragmentation and repeated collection is improved, forming a continuous and traceable medical record from pre-diagnosis to recovery.

[0015] Secondly, based on the diagnostic results and triage level, a disease state machine is instantiated to automatically generate a follow-up plan that matches the child's condition. At the same time, by calculating the deviation of the condition, the follow-up frequency and content are dynamically adjusted in a step-by-step manner, so that post-diagnosis management is transformed from a fixed mode to a data-driven adaptive mode.

[0016] Third, by using the event tracing model, the data of the entire course of the disease is aggregated into a health record with the patient's ID as the primary key according to the time dimension. This record fully documents the symptom evolution, diagnosis and treatment conclusions and outcomes of each fever event, providing a reliable data foundation for subsequent diagnosis and treatment and epidemiological analysis. Attached Figure Description

[0017] Figure 1 This is a flowchart of a method for managing fever throughout the entire course of childhood, as provided in this application.

[0018] Figure 2 This is a flowchart of the method for generating corresponding post-diagnosis follow-up plans based on diagnostic results and triage levels, provided in this application.

[0019] Figure 3 This application provides a flowchart of a comprehensive management system for the entire course of fever in children. Detailed Implementation

[0020] To better understand the purpose, technical solutions, and advantages of this application, it has been described and illustrated below with reference to the accompanying drawings and embodiments. However, those skilled in the art should understand that this application can be implemented without these details. It will be apparent to those skilled in the art that various modifications can be made to the embodiments disclosed in this application, and the general principles defined in this application can be applied to other embodiments and application scenarios without departing from the principles and scope of this application. Therefore, this application is not limited to the illustrated embodiments, but is consistent with the broadest scope claimed in this application.

[0021] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0022] Figure 1 This is a flowchart illustrating a method for managing fever throughout the entire course of childhood illness, as provided in this application. Figure 1 As shown, a method for managing fever throughout the entire course of a child's illness includes the following steps: Step S100: Obtain multi-turn dialogue information related to the child's fever, identify danger signals based on the multi-turn dialogue information, and generate triage results including a medical history summary and triage level.

[0023] This application describes the process from the management perspective. First, it requires acquiring multi-round dialogue information related to the child's fever. Specifically, this involves collecting the dialogue text input by the parent through a mobile interface. The management system pre-configures a standard diagnostic framework for children's fever, breaking down the data collection process into a chief complaint layer, a deeper understanding of the present illness, a comprehensive review of accompanying symptoms, and a layer for anchoring past medical history and personal history. A dynamic pruning decision tree engine and a medical natural language understanding engine work collaboratively. Each round of questions is dynamically generated based on the parent's previous response. The medical natural language understanding engine extracts positive symptom entities from the response text, and the decision tree engine automatically determines the direction of the next round of follow-up questions or prunes the branch based on the entity status, until the information density of all branches reaches the required level or the parent actively ends the data collection process. Dynamic pruning shortens ineffective inquiry paths and lowers the barrier to medical expression for parents, allowing the management system to complete information collection without requiring parents to possess medical knowledge.

[0024] Based on the completion of multi-round dialogue information collection, further structured analysis and risk quantification assessment of the collected multi-source information are required to output standardized triage results and medical history summaries. The process of identifying warning signals and generating triage results including medical history summaries and triage levels based on multi-round dialogue information includes the following steps: Step S101: Obtain the response text and fever-related time-series data from the multi-round dialogue information, extract positive symptom entities from the response text layer by layer according to the preset consultation skeleton, and generate the body temperature change trend based on the fever-related time-series data.

[0025] Step S102: Based on the positive symptom entities and the temperature change trend, update the danger signal flag bit vector according to the danger signal list, and record the first occurrence time and confidence level of each danger signal.

[0026] Step S103: Calculate the baseline score, the weighted value of the danger signal, and the uncertainty penalty value based on the positive symptom entity, the trend of body temperature change, the danger signal marker vector, and the confidence level. Then, sum the baseline score, the weighted value of the danger signal, and the uncertainty penalty value to obtain the urgency score.

[0027] Step S104: Match the corresponding triage level according to the urgency score, and combine the positive symptom entities and body temperature change trends extracted in each round into a medical history summary according to the time series.

[0028] The management system pre-configures a standard questionnaire framework for pediatric fever consultation. This framework is not a fixed questionnaire, but rather a logical structure that breaks down the data collection process into four layers: chief complaint collection, in-depth present medical history, accompanying symptom traversal, and anchoring of past medical history and personal history. The management system uses a mobile interface to ask parents questions in a multi-round dialogue format and obtains their input text responses. Each round of questions is dynamically generated based on the previous round's responses. A medical natural language understanding (NLU) engine is used to extract structured clinical information, namely positive symptom entities such as fever, cough, and vomiting, from the parents' responses. Simultaneously, it also obtains time-series data related to fever input by the parents, such as body temperature measurements at multiple time points, and generates a description of the temperature change trend over time based on this data.

[0029] After each round of dialogue information is parsed, a multi-dimensional feature accumulation vector is updated synchronously. This vector contains a danger signal flag vector. The management end refers to a predefined list of fever danger signals in children, such as lethargy and difficulty waking up, persistent seizures, and rapid breathing, and matches the extracted positive symptom entities and body temperature change trends with the rules in the list. Once a match is successful, it indicates that a danger signal has been triggered, and the corresponding position of the danger signal in the flag vector is set, for example, the flag is changed from 0 to 1, and the first occurrence time of the signal and the confidence level of this match are recorded simultaneously.

[0030] Then, a three-layer calculation logic of basic score + uncertainty penalty for danger signals is used to determine the urgency score. The first layer is the basic score, which is a lookup score based on objective indicators. That is, based on objective indicators such as peak body temperature, duration of fever, and age of the child obtained from positive symptoms and temperature change trends, a basic score is obtained by looking up a table in a preset scoring table. For example, the higher the peak body temperature, the longer the duration of fever, and the younger the child, the higher the basic score. The second layer is the danger signal weighting, which is a response to the danger signal flag bit vector. That is, the danger signal flag bit vector is checked, and for any flag that is set, its corresponding preset weight is obtained and superimposed on the basic score in a weighted manner. In particular, when multiple danger signals exist at the same time, their comprehensive weighting coefficients are superimposed in a superlinear manner to reflect the synergistic risk of multiple risk factors coexisting. The third layer is the uncertainty penalty, which is a consideration of information quality. This involves assessing the confidence level of responses to key information such as onset time, highest body temperature, and mental state. If the confidence level of multiple key information items is below a preset threshold, it indicates that the information on which the current assessment is based is unclear. Therefore, for safety reasons, the management automatically adds an uncertainty penalty value, thus increasing the final score. Finally, the scores of the above three items are added together to obtain a comprehensive urgency score.

[0031] The management system pre-divides multiple consecutive scoring intervals, each corresponding to a triage level, such as four levels: normal observation, recommended medical attention, prompt medical attention, and immediate emergency care. After calculating the urgency score, it determines which preset interval the score falls into, thus matching the corresponding triage level. Simultaneously, a summary generator is activated, arranging and combining the positive symptom entities extracted from each round of the consultation process, such as fever, cough, and runny nose, and the temperature change trend composed of temperature values, in chronological order with natural language to generate a clearly structured and logically coherent medical history summary. This summary, together with the final triage level, constitutes the triage result, providing a structured data foundation for subsequent medical procedures.

[0032] By combining dynamic decision trees with natural language understanding and introducing a multi-dimensional quantitative scoring mechanism, this method achieves structured data collection and objective grading of children's fever symptoms. It effectively addresses issues such as fixed consultation paths, inaccurate information collection, and a lack of objective quantitative grading, improving the accuracy, security, and data availability of pre-diagnosis triage and providing a reliable foundation for building seamless data flow from pre-diagnosis to diagnosis.

[0033] Step S200: During the consultation, the triage results are synchronized to the doctor's end, and the diagnosis results input by the doctor based on the triage results are obtained.

[0034] Upon arrival at the hospital, the management system matches the child's identity using their ID card or medical card number and automatically pushes the generated triage results to the corresponding doctor's workstation in the HIS system. The doctor's workstation serves as the front-end interface for the electronic medical record system, allowing doctors to directly view the triage results during consultations. Doctors confirm the diagnosis based on the medical history summary and triage level, combined with the child's real-time vital signs. If the doctor determines the triage result is accurate, they confirm it with a single click and populate it into the electronic medical record. If there are discrepancies, the doctor can directly modify the medical history summary and triage level in the interface using revision mode. The management system stores the differences before and after the modification as field-level modification records under the current fever event ID. After completing the diagnosis, the doctor enters the diagnostic results, including disease type and severity classification, through the doctor's workstation and associates the diagnostic results with the current fever event ID for post-diagnosis retrieval. This seamlessly integrates pre-diagnosis data into the diagnosis process, effectively reducing duplicate consultations and improving doctors' diagnostic efficiency.

[0035] Step S300: Generate a corresponding post-diagnosis follow-up plan based on the diagnosis results and triage level.

[0036] After obtaining the diagnosis results entered by the doctor, the post-diagnosis stage begins. This stage uses the diagnosis results and triage level as input, and automatically generates a post-diagnosis follow-up plan tailored to the individual child's condition using a pre-set disease state machine template. Generating the corresponding post-diagnosis follow-up plan based on the diagnosis results and triage level includes the following steps: Step S301: Obtain the disease type and severity grade from the diagnostic results, as well as the age risk level of the child represented by the triage level. Input the disease type, severity grade, and age risk level of the child into a preset pediatric fever course state machine template to instantiate and generate an individualized state machine that includes the acute fever phase, fever reduction observation period, recovery period, symptom recurrence pending judgment period, deterioration warning period, and event closure period, as well as the state transition conditions.

[0037] Step S302: Generate nursing content identifiers and self-assessment questionnaire identifiers corresponding to each preset follow-up contact point according to the preset duration of each state in the individualized state machine and the preset follow-up contact points, and configure a post-diagnosis follow-up plan according to the time sequence of each preset follow-up contact point and the corresponding nursing content identifiers and self-assessment questionnaire identifiers.

[0038] The management system has a pre-set state machine template for the course of a child's fever. This template defines six core states that a child may experience during the course of a fever: acute fever phase, fever reduction observation period, recovery period, symptom recurrence and pending judgment period, deterioration warning period, and event closure period, as well as the state transition conditions between each state. These state transition conditions are pre-set in the template as rules, such as a transition from the fever reduction observation period to the recovery period if the body temperature is normal for 48 consecutive hours and the main symptoms disappear, and a transition from the recovery period or the fever reduction observation period to the symptom recurrence and pending judgment period if the body temperature rises again to above 38.5°C. The management system parses the disease type and severity grade entered by the doctor from the diagnostic results, and extracts the child's age-risk level represented by the triage level from the triage results. By inputting the above disease type, severity grade, and child's age-risk level as instantiation parameters into the pre-set state machine template for the course of a child's fever, an individualized state machine matching the fever event can be instantiated. This personalized state machine includes the above six core states and their corresponding state transition conditions, but the duration and follow-up contact points configured for each state in subsequent steps will be personalized according to the disease type, severity level and the child's age risk level.

[0039] Each state in the individualized state machine is pre-configured with a default duration and planned follow-up touchpoints associated with that state. For example, in the individualized state machine, the default duration of the acute fever phase is 3 days, and the planned follow-up touchpoints are set for days 1 and 3 of the illness; the default duration of the fever reduction observation period is 2 days, and the planned follow-up touchpoint is day 5 of the illness; the default duration of the recovery period is 7 days, and the planned follow-up touchpoints are day 7 and 14 of the illness. The management terminal presets corresponding nursing content identifiers and self-assessment questionnaire identifiers for each planned follow-up touchpoint. The nursing content identifier points to specific nursing items in the pre-set nursing guidance content library, such as physical cooling guidance, nebulized inhalation care, and methods of chest percussion and sputum expectoration. The self-assessment questionnaire identifier points to specific questionnaire templates in the pre-set symptom self-assessment questionnaire library, such as questionnaire templates containing items such as cough frequency score, sputum characteristics, and respiratory rate observation. The management system determines the specific time points for each preset follow-up touchpoint based on the preset duration of each state in the individualized state machine. It then obtains the corresponding nursing content identifiers and self-assessment questionnaire identifiers for each preset follow-up touchpoint. Finally, it arranges and combines these preset follow-up touchpoints and their corresponding nursing content identifiers and self-assessment questionnaire identifiers in chronological order to form a complete post-diagnosis follow-up plan. This plan includes the specific trigger time points, the nursing guidance content identifiers to be pushed at each time point, and the symptom self-assessment questionnaire identifiers. On one hand, the state machine instantiation driven by diagnostic results and triage levels ensures that the content and pace of the follow-up plan are precisely matched to the individual child's condition. On the other hand, state transition conditions enable automatic responses to changes in disease progression, allowing the follow-up plan to dynamically adjust as the condition evolves. Furthermore, the preset nursing content identifiers and self-assessment questionnaire identifiers enable differentiated nursing guidance and symptom tracking for different disease types and severity levels, improving the personalization and efficiency of post-diagnosis follow-up.

[0040] Step S400: In the post-diagnosis stage, obtain the follow-up data reported based on the post-diagnosis follow-up plan, and calculate the deviation of the condition based on the follow-up data, medical history summary and diagnosis results.

[0041] The post-diagnosis phase then begins. Based on the time sequence of the preset follow-up touchpoints included in the post-diagnosis follow-up plan generated in step S300, the management terminal pushes follow-up task reminders to the child's parents via a preset multi-channel push interface when each preset follow-up touchpoint is reached. This multi-channel push interface includes at least one of an APP push gateway, a WeChat service account gateway, and an SMS gateway. The follow-up task reminder includes a nursing content identifier and a self-assessment questionnaire identifier corresponding to the current preset follow-up touchpoint. Responding to the follow-up task reminder, the parents view the nursing guidance content corresponding to the nursing content identifier through the mobile interface and complete the symptom self-assessment questionnaire corresponding to the self-assessment questionnaire identifier. The management terminal receives the questionnaire data reported by the parents through the mobile interface, and also receives temperature measurement data and abnormal manifestation text proactively reported by the parents. The management terminal performs structured parsing on the received questionnaire data, extracting symptom quantitative scores and symptom entities; it performs outlier removal and time alignment processing on the received temperature measurement data to generate the currently reported temperature time-series data; and it uses a medical natural language understanding engine to extract newly occurring symptom entities from the received abnormal manifestation text. The management system associates processed questionnaire data, temperature measurement data, and abnormal symptom text by reporting timestamp and encapsulates them using the current fever event ID as the primary key, forming structured follow-up data. This follow-up data includes timestamps, symptom quantitative scoring sequences, symptom entity lists, temperature time-series data, and lists of newly occurring symptom entities. Through multi-channel push notifications and structured data collection, the system ensures the completeness, timeliness, and structure of the follow-up data, providing high-quality data input for accurate calculation of subsequent disease deviations.

[0042] Figure 2 This is a flowchart illustrating the method for calculating the degree of deviation of the disease condition based on follow-up data, medical history summary, and diagnostic results, as provided in this application. Figure 2 As shown, calculating the deviation of the condition based on follow-up data, medical history summary, and diagnostic results includes the following steps: Step S401: Extract body temperature time series data, symptom quantitative scores and new symptom entities from follow-up data, extract historical body temperature peaks from medical history summary, and match the corresponding expected recovery curve and expected symptom improvement trajectory for the same disease from the diagnosis results.

[0043] Step S402: Calculate the dynamic time curvature distance between the body temperature time series data and the expected recovery curve of the same disease, and normalize the dynamic time curvature distance to obtain the time series deviation.

[0044] Step S403: Calculate the absolute value of the residual between the symptom quantitative score and the expected symptom improvement trajectory at the same time point. After weighted summation of the absolute values ​​of the residuals at each time point, normalize to obtain the symptom outcome deviation.

[0045] Step S404: Match the newly emerging symptom entity with the preset list of danger signals. When the match is successful, calculate the penalty value of the newly emerging danger signal according to the preset weight of the matched danger signal. When the match fails, set the penalty value of the newly emerging danger signal to zero.

[0046] Step S405: The disease deviation is obtained by weighted summation of the time sequence deviation, symptom outcome deviation, and new danger signal penalty value.

[0047] The management system extracts three core data dimensions from the packaged follow-up data: first, time-series temperature data, which is a sequence of temperature values ​​reported by the child at different time points after diagnosis, arranged chronologically; second, symptom quantitative scoring, which is a scoring sequence formed by semi-quantitatively scoring the severity of key symptoms in the symptom self-assessment questionnaires filled out by parents at each follow-up contact point; and third, newly occurring symptom entities, which are symptom entities that did not appear before or during the diagnosis, extracted from the abnormal manifestation text reported by parents through the NLU engine. Simultaneously, the management system retrieves the historical peak temperature recorded before the diagnosis for this fever event from the medical history summary generated before diagnosis, as a reference benchmark for temperature recovery assessment. Furthermore, the management system uses the disease type and severity grade in the diagnosis results as an index to retrieve the expected recovery curve and expected symptom improvement trajectory for the same disease and severity grade as the child from a pre-set recovery trajectory template library. The expected recovery curve is a standardized reference curve representing the gradual decrease in body temperature to the normal range over time during the course of the disease; the expected symptom improvement trajectory is a standardized reference trajectory representing the gradual decrease in the quantitative score of key symptoms to the level of recovery over time during the course of the disease. Both expected trajectories are derived from statistical fitting of similar historical cases and represent the typical recovery process of this type of disease under standard prognosis.

[0048] Because the reported body temperature time points of the children cannot be perfectly aligned with the time axis of the expected recovery curve, and because there are slight deviations in the time axis of body temperature recovery rhythm among different individuals, traditional point-by-point difference comparison cannot objectively reflect the similarity of the two curves in terms of morphology. Therefore, a dynamic time bending algorithm is used to calculate the dynamic time bending distance between the body temperature time series data and the expected recovery curve of the same disease. The dynamic time bending distance is calculated by non-linearly aligning the two sequences on their time axes to find the optimal bending path that minimizes the cumulative distance between points in the two sequences. The smaller this distance value, the closer the morphology of the actual body temperature recovery process is to the expected recovery process. Furthermore, since this distance is an absolute quantity, its value is affected by the range of body temperature values ​​and the sequence length, and cannot be directly compared between different cases. Therefore, the management divides the dynamic time bending distance by a preset normalized baseline value, which is set to twice the statistical median of the dynamic time bending distance in the historical data of the same disease. This maps the dynamic time bending distance to a dimensionless value between 0 and 1, obtaining the time series deviation. The closer the time series deviation is to 0, the more the child's body temperature recovery process conforms to the typical recovery pattern of this type of disease; the closer it is to 1, the more significantly the body temperature recovery process deviates from the typical pattern.

[0049] For the quantitative assessment of symptom recovery, the management system calculates the absolute difference between the quantitative symptom scores reported at each follow-up point and the expected scores at the corresponding time points in the expected symptom improvement trajectory, obtaining the absolute residual value at each time-series location. Considering the different importance of different symptoms in the disease assessment, different weighting coefficients are preset for each symptom dimension. For example, for cough and sputum-related diseases, the cough frequency score has a higher weight than the appetite score. The management system multiplies the absolute residual values ​​of all symptom dimensions at each time-series location by their respective weighting coefficients and sums them to obtain the weighted residual sum of each time-series location. Then, the weighted residual sums of each time-series location are arranged in chronological order to form a residual sequence, and the mean of this residual sequence is calculated. Dividing this mean by a preset normalized benchmark value yields the symptom outcome deviation, which is set to twice the median of the weighted residual mean in historical data for the same disease. The symptom outcome deviation is also a dimensionless value between 0 and 1. The smaller the value, the more the symptom recovery process conforms to the typical improvement trajectory, and the larger the value, the more the symptom recovery deviates from the expected trajectory.

[0050] In the post-diagnosis phase, children may exhibit new warning signs that were not present before the diagnosis. These signs cannot be adequately reflected by temperature curves or routine symptom scores. The management system matches the newly emerging symptom entities extracted in step S401 with a pre-set list of fever warning signs in children. If any newly emerging symptom entity falls into the warning sign list, it indicates a new warning sign has appeared in the post-diagnosis phase. The preset weight value of the matched warning sign in the pre-set warning sign weight table is obtained, and this preset weight value is used as the penalty value for the newly emerging warning sign. If multiple warning signs are matched simultaneously, the preset weight values ​​of each warning sign are added together, and the sum is multiplied by a synergistic factor such as 1.5 to represent the synergistic risk of multiple newly emerging warning signs coexisting. If none of the newly emerging symptom entities are matched with the warning sign list, the penalty value for the newly emerging warning sign is set to zero.

[0051] Finally, the obtained temporal deviation, symptom outcome deviation, and new danger signal penalty value are weighted and summed. The weights of temporal deviation, symptom outcome deviation, and new danger signal penalty value are set according to a preset weighting scheme, and the sum of their weights is 1. Preferably, the weights are 0.4 for temporal deviation, 0.4 for symptom outcome deviation, and 0.2 for new danger signal penalty value. The resulting weighted sum is a dimensionless comprehensive score used to quantitatively characterize the overall deviation between the child's current actual recovery process and the typical expected recovery process for this type of disease. Thus, by quantifying and weighting the multidimensional deviations of body temperature temporal data, symptom quantitative scores, and new symptom entities, the problem of single-point indicators being unable to characterize the deviation in disease progression is improved, enabling an objective, quantitative, and real-time assessment of the deviation between the child's recovery trajectory and the expected path. The deviation of the condition provides precise data support for the dynamic adjustment of the follow-up plan and the early warning of deterioration, enabling post-diagnosis management to transform from a fixed model to a data-driven adaptive model, thereby improving the accuracy and timeliness of post-diagnosis management.

[0052] Step S500: When the deviation of the condition reaches the preset trigger condition, adjust the post-diagnosis follow-up plan and issue an early warning.

[0053] After calculating the deviation rate of the patient's condition, the management system immediately enters the deviation response phase, driving dynamic adjustments and early warning triggers to the follow-up plan based on the quantitative results of the deviation rate. Specifically, adjusting the post-diagnosis follow-up plan when the deviation rate reaches a preset trigger condition includes the following steps: Step S501: Compare the degree of deviation of the condition with the preset multi-level deviation threshold to obtain the deviation level.

[0054] Step S502: When the deviation level is mild, the post-diagnosis follow-up plan remains unchanged.

[0055] Step S503: When the deviation level is moderate, insert a temporary follow-up contact between two adjacent preset follow-up contacts in the post-diagnosis follow-up plan, and push close observation guidance content to the temporary follow-up contact.

[0056] Step S504: When the deviation level is significant, the individualized state machine is forcibly migrated to the deterioration warning period. Based on the deterioration warning period, the preset follow-up contacts that are not currently executed in the post-diagnosis follow-up plan are suspended, and the high-frequency monitoring scheme is started.

[0057] The management system has a pre-defined multi-level deviation threshold system, including a mild deviation threshold T1 and a significant deviation threshold T2, where T1 < T2. The values ​​of T1 and T2 are determined by percentile statistics of the deviation distribution of historical cases of the same disease, with T1 set at the 50th percentile and T2 at the 85th percentile. The management system compares the obtained deviation degree with T1 and T2 respectively to determine the current deviation level of the child. When the deviation degree is ≤ T1, it is considered mild deviation; when T1 < deviation degree ≤ T2, it is considered moderate deviation; and when the deviation degree is > T2, it is considered significant deviation. By mapping continuous values ​​to discrete levels, a clear basis for branching judgments is provided for subsequent graded responses.

[0058] When the deviation level is determined to be mild, it indicates that the child's recovery process is within an acceptable range compared to the typical recovery process for similar diseases, and has not yet reached the point where intervention is required. The management system maintains all configurations of the current post-diagnosis follow-up plan unchanged, including the time schedule for each preset follow-up touchpoint, the corresponding nursing content identifiers and self-assessment questionnaire identifiers for each touchpoint, and the current state and state transition conditions of the individualized state machine. Follow-up tasks continue to be pushed according to the original post-diagnosis follow-up plan schedule, without increasing the follow-up frequency.

[0059] When the deviation level is determined to be moderate, it indicates that the child's recovery process has deviated significantly, requiring increased monitoring intensity but not yet reaching the level of emergency intervention. The management system initiates an enhanced follow-up frequency protocol. This involves identifying a gap between two adjacent pre-set follow-up touchpoints on the post-diagnosis follow-up plan timeline after the current time point and inserting a temporary follow-up touchpoint at the midpoint of this gap. The temporary follow-up touchpoint pushes content including a close observation guidance content identifier retrieved from the nursing guidance content library corresponding to the current disease stage. This close observation guidance content points to nursing items including key points for identifying disease aggravation, indications for re-visiting medical attention, and enhanced home care measures. Simultaneously, an enhanced self-assessment questionnaire identifier matching the current disease type is retrieved from the symptom self-assessment questionnaire library. This enhanced self-assessment questionnaire, compared to the regular questionnaire, adds requirements for the frequency of temperature measurement and daily comparison items for key symptoms. The management system, according to the chronological order of the temporary follow-up touchpoint, includes the temporary follow-up touchpoint and its corresponding nursing content identifier and self-assessment questionnaire identifier in the execution queue of the post-diagnosis follow-up plan. Aside from inserting temporary follow-up contact points, the existing preset follow-up contact points and their corresponding push notifications remain unchanged. If the disease deviation rate continues to be in the moderate deviation range in subsequent calculations, the management system can repeat the operation of inserting temporary follow-up contact points in new gaps to achieve a step-by-step increase in monitoring frequency.

[0060] When the deviation level is determined to be significant, it indicates that the child's recovery process has seriously deviated from the typical trajectory, posing a high risk of disease deterioration. The management system performs a forced state transition, ignoring the current state of the individualized state machine and its preset state transition conditions, directly forcibly transitioning the individualized state machine from its current state to the deterioration warning period. The deterioration warning period is one of the six core states preset in the child's fever course state machine template. When the individualized state machine enters this state, the management system reads the corresponding state configuration rules. The state configuration rules stipulate that during the deterioration warning period, all preset follow-up touchpoints in the post-diagnosis follow-up plan that have not yet been executed, along with their corresponding nursing content pushes and self-assessment questionnaire pushes, are suspended. Simultaneously, a high-frequency monitoring program was initiated. This program involves switching the follow-up data collection frequency from the original trigger-based mode to a daily high-frequency collection mode. During each high-frequency collection, a critical care version of the self-assessment questionnaire matching the current disease type is retrieved and pushed to the system. Compared to the regular and enhanced versions, this version adds items for dyspnea scoring, consciousness status scoring, and observation of dehydration signs. The reported follow-up data is input in real-time into the disease deviation calculation process in step S400, achieving daily rolling updates of the disease deviation. The high-frequency monitoring program continues until the individualized state machine transitions out of the deterioration warning period through state transition conditions. Through the above-mentioned tiered response mechanism, the follow-up plan is adaptively adjusted. When the disease deviation is mild, the original plan is maintained to avoid over-intervention; when it is moderate, temporary follow-up triggers are inserted to increase monitoring density; when it is significant, a forced state transition suspends regular follow-up and initiates the high-frequency monitoring program, forming a gradient response strategy that matches the degree of disease deviation.

[0061] It should be understood that although steps S501 to S504 are described sequentially in this application, the execution of these steps does not necessarily have to follow this order. After the deviation degree of the condition is calculated, steps S501 to S504 are executed in the described logical order, that is, first the deviation level is determined, and then the corresponding adjustment strategy is selected according to the deviation level. The data processing flow involved in the above specific embodiments can be executed centrally on the management server side, or it can be distributed between the cloud server and edge nodes according to the system architecture design, which is not limited here.

[0062] Preferably, after the high-frequency monitoring program is activated, the medical history summary, diagnosis results and reported follow-up data corresponding to the current fever event are retrieved, and the disease retrospective summary is generated by merging them according to the preset retrospective summary template. The disease retrospective summary is then pushed to the bound doctor's terminal.

[0063] Specifically, the management system uses the current fever event ID as an index to retrieve the medical history summary generated before diagnosis, the diagnostic results entered by the doctor during diagnosis, and all follow-up data reported after diagnosis from the event storage. The management system then reorganizes this data according to a preset retrospective summary template. This template includes sections such as event overview, diagnostic conclusion, key disease progression nodes (first fever time, peak temperature, and records of warning signal appearance and disappearance), current status (latest temperature, symptom score, and list of executed follow-up points), and reasons for triggering deterioration warnings. This integrates the structured data scattered throughout the entire disease progression into a coherent disease retrospective summary, which is then pushed to the doctor associated with the child. This allows the doctor to quickly grasp the complete evolution of the child's condition since admission, enabling timely intervention decisions. By generating a disease retrospective summary, doctors can quickly obtain a condensed and focused overview of the disease's evolution when the child's condition significantly deviates from expectations, without having to painstakingly review the original records scattered across different stages, thus improving the efficiency of clinical response in high-risk situations.

[0064] Step S600: Generate health records based on fever events, using multi-round dialogue information, triage results, diagnosis results, and follow-up data.

[0065] After completing the post-diagnosis stage of disease monitoring, deviation calculation, and graded early warning adjustment, in order to integrate the fragmented data scattered across the pre-diagnosis, diagnosis, and post-diagnosis stages into a continuous and traceable structured disease record, it is necessary to further construct health records based on fever events. The construction process of health records not only serves the historical retrospective of current fever events but also provides a standardized data foundation for the subsequent assessment of similar diseases and regional epidemiological analysis through an event tracing model. Unlike the traditional hospital information system (HIS) which uses single-visit records as the basic unit and stores data in a scattered manner with difficulty in cross-stage correlation, this application adopts an event tracing model in its underlying data architecture. Instead of simply overwriting the latest data generated in each stage (pre-diagnosis, diagnosis, and post-diagnosis), it persistently stores each interaction and state change as an immutable event. Generating health records based on multi-turn dialogue information, triage results, diagnostic results, and follow-up data, using fever events as units, includes the following steps: Step S601: Write the multi-round dialogue information, triage results, diagnosis results and follow-up data into the event log indexed by the fever event ID of the fever event in chronological order to obtain the event log stream.

[0066] Step S602: Aggregate the event log stream by time dimension into a materialized view with the child ID as the primary key to obtain a health record. Each fever event record in the health record includes the symptom evolution map corresponding to the multi-turn dialogue information, the urgency change log corresponding to the triage result, the diagnosis conclusion corresponding to the diagnosis result, and the body temperature curve and outcome status corresponding to the follow-up data.

[0067] First, using a globally unique fever event ID as the aggregation root identifier, all data generated throughout the entire course of this fever event is appended to the event log storage area in chronological order of occurrence. The data written includes, but is not limited to, the original question-and-answer pairs for each round of dialogue in the pre-diagnosis stage, snapshot records of each multi-dimensional feature accumulation vector, change logs for each urgency score, and the medical history summary generated before diagnosis; version snapshots of pre-filled medical records in the diagnosis stage, difference records of each field-level modification operation by the doctor, and the final diagnosis results and prescription information entered by the doctor; and each follow-up data reported by parents in the post-diagnosis stage, the calculation results of each deviation in the condition, each state transition event of the individualized state machine, and each warning trigger record. Each piece of data is written with a precise timestamp and operation source identifier, and is stored in the event log in an append-only manner. Log entries already written are not modified or deleted. When data correction is needed, a new correction record is appended to express the change, forming an immutable event log stream that completely preserves the entire historical trajectory of the fever event from pre-diagnosis to post-diagnosis.

[0068] Using the child's ID as the primary key, the event log streams corresponding to each of the child's fever events are aggregated and projected along the time dimension to construct a materialized view of the child, i.e., a health record. During this construction process, the raw event data in the event log stream is cleaned, standardized in format, and aligned along the time axis to remove redundant and duplicate data. Data scattered across different locations in the log stream under the same fever event ID is reorganized chronologically. For multi-round dialogue information in the pre-diagnosis stage, the management end extracts the symptom onset time, resolution time, and severity changes from the positive symptom entities and temperature measurements extracted in each round, combining them to form a symptom evolution map describing the evolution of symptoms over time for this fever event. For triage results in the pre-diagnosis stage, the management end extracts the score values ​​and corresponding triage levels at each time point from the urgency score change log, forming an urgency change log describing the dynamic changes in the urgency of this fever event. For diagnostic results in the diagnosis stage, the management end directly saves the diagnostic conclusion and corresponding medical orders. For follow-up data after diagnosis, all reported temperature values ​​were fitted into a complete temperature curve in chronological order, and the final outcome was recorded. Each fever event record in the health record was arranged in chronological order, with each fever event of the same child presented in reverse chronological order of onset time, allowing doctors to quickly review the complete course of the child's fever.

[0069] Preferably, this application also obtains the pre-diagnosis triage direction and the final diagnosis during diagnosis for the same fever event in the health record, performs structured alignment on the pre-diagnosis triage direction and the final diagnosis during diagnosis, and extracts the difference features to obtain a diagnostic deviation set; generates a correction feedback package containing field-level modified records and diagnostic direction deviations based on the diagnostic deviation set, and sends the correction feedback package to the pre-diagnosis consultation engine to update the pruning threshold and follow-up question trigger weight of the decision tree node.

[0070] After generating health records based on fever events, a feedback loop from pre-diagnosis consultation to in-diagnosis diagnosis is constructed by verifying the consistency of data across different stages of the same fever event within the health record. Specifically, using the fever event ID as an index, the pre-diagnosis triage direction and the final in-diagnosis diagnosis corresponding to the same fever event are retrieved from the generated health records. The pre-diagnosis triage direction is the suspected disease direction summarized by the decision tree path in the pre-diagnosis stage, stored in the initial metadata object of the fever event in the form of structured tags, such as "high probability of viral upper respiratory tract infection" and "lower respiratory tract infection to be ruled out." The final in-diagnosis diagnosis is the final disease type entered by the doctor in the in-diagnosis stage, such as "bronchopneumonia." The pre-diagnosis triage direction and the final in-diagnosis diagnosis are structurally aligned according to a preset clinical classification system, that is, the corresponding node positions of the two are located on the node tree of the classification system. Then, the differences between the two are extracted, including but not limited to whether the triage direction and the final diagnosis are consistent, the classification system node level and deviation direction involved when they are inconsistent, and the doctor's modification operation records of each medical history field in the pre-filled medical record in the in-diagnosis stage. The above-mentioned differences are summarized to form a diagnostic bias set.

[0071] The diagnostic deviation set is then encapsulated into a standardized correction feedback package according to a pre-defined feedback package structure. The correction feedback package includes at least the fever event ID, a set of differential features, decision tree node path backtracking markers, and key clinical variables before and after modification. The decision tree node path backtracking markers point to the sequence of decision tree nodes that triggered the current triage direction during the pre-diagnosis consultation, identifying the information path upon which this triage was based. Key clinical variables include positive symptom entities and physical sign data related to the diagnostic deviation collected during the pre-diagnosis consultation. The correction feedback package is not directly used to modify any data of the current fever event; instead, after desensitization, it is sent in batches to the incremental update queue of the pre-diagnosis consultation engine according to time windows. The pre-diagnosis consultation engine uses the accumulated correction feedback packages as input and periodically updates the decision tree parameters. For decision paths that frequently appear in the diagnostic deviation set, the pruning threshold corresponding to that path is lowered, making it more difficult to prune that branch in subsequent consultations, thereby collecting more information. Simultaneously, the trigger weight for follow-up questions corresponding to symptom entities that are repeatedly refined or modified by doctors in the differential features is increased, making follow-up questions related to that symptom more prioritized in subsequent consultations. Through the aforementioned feedback mechanism, the diagnostic conclusions generated during the diagnosis phase correct the pre-diagnosis inquiry strategy, ensuring that the data collection path of the decision tree continuously converges with the actual clinical diagnostic logic. By quantifying the deviation between the diagnosis conclusions during diagnosis and the pre-diagnosis triage direction into structured feedback signals, and using them to dynamically adjust the pruning threshold and follow-up question weights of the pre-diagnosis decision tree, a closed-loop feedback loop from post-diagnosis to pre-diagnosis is achieved. This allows for continuous self-optimization of the entire disease management process, improving the consistency between the pre-diagnosis triage direction and the final clinical diagnosis.

[0072] Figure 3 This is a block diagram of a comprehensive management system for childhood fever, as provided in this application. Figure 3 As shown, a comprehensive fever management system for children includes: an identification module for acquiring multi-turn dialogue information related to fever, identifying warning signals based on the multi-turn dialogue information, and generating triage results including a medical history summary and triage level; a consultation module for synchronizing the triage results to the doctor's end during consultation and acquiring the diagnosis results input by the doctor based on the triage results; a follow-up module for generating a corresponding post-diagnosis follow-up plan based on the diagnosis results and triage level, acquiring follow-up data reported based on the post-diagnosis follow-up plan during the post-diagnosis stage, and calculating the deviation degree of the condition based on the follow-up data, medical history summary, and diagnosis results; an early warning module for adjusting the post-diagnosis follow-up plan and issuing an early warning when the deviation degree of the condition reaches a preset trigger condition; and a management module for generating health records based on fever events, using multi-turn dialogue information, triage results, diagnosis results, and follow-up data as units.

[0073] The other functions performed by the aforementioned identification module, medical treatment module, follow-up module, early warning module, and management module, as well as the technical details of each function, are the same as or similar to the corresponding features in the previously described method for managing the entire course of childhood fever, and therefore will not be repeated here.

[0074] This application also provides a computer storage medium storing a computer program that, when run on a computer, enables the computer to execute the steps in the previously described method for managing the entire course of fever in children.

[0075] It should be understood that although the steps in the flowcharts in the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily performed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order requirement for the execution of these steps, and they can be performed in other orders.

[0076] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for managing fever throughout the entire course of childhood illness, characterized in that, The method includes: Acquire multi-turn dialogue information related to fever in children, identify danger signals based on the multi-turn dialogue information, and generate triage results including medical history summary and triage level; During a medical visit, the triage results are synchronized to the doctor's end, and the diagnosis results input by the doctor based on the triage results are obtained; A corresponding post-diagnosis follow-up plan is generated based on the diagnostic results and the triage level; In the post-diagnosis stage, follow-up data reported based on the post-diagnosis follow-up plan is obtained, and the deviation of the condition is calculated based on the follow-up data, the medical history summary, and the diagnosis results. When the deviation of the condition reaches the preset trigger condition, the post-diagnosis follow-up plan is adjusted and an early warning is issued. Based on the multi-round dialogue information, the triage results, the diagnosis results, and the follow-up data, a health record is generated with fever events as the unit.

2. The method according to claim 1, characterized in that, The process of identifying danger signals based on the multi-turn dialogue information and generating triage results including a medical history summary and triage level includes: The system acquires the response text and fever-related time-series data from the multi-turn dialogue information, extracts positive symptom entities from the response text layer by layer according to the preset consultation skeleton, and generates a body temperature change trend based on the fever-related time-series data. Based on the positive symptom entities and the temperature change trend, update the danger signal flag bit vector according to the danger signal list, and record the first occurrence time and confidence level of each danger signal; Based on the positive symptom entities, the body temperature change trend, the danger signal marker vector, and the confidence level, a base score, a danger signal weighted value, and an uncertainty penalty value are calculated. The base score, the danger signal weighted value, and the uncertainty penalty value are then superimposed to obtain an urgency score. Based on the urgency score, the corresponding triage level is matched, and the positive symptom entities and body temperature change trends extracted in each round are combined in time series to form a medical history summary.

3. The method according to claim 2, characterized in that, The process of generating a corresponding post-diagnosis follow-up plan based on the diagnostic results and the triage level includes: Obtain the disease type and severity grade from the diagnostic results, as well as the child's age risk level represented by the triage level. Input the disease type, the severity grade, and the child's age risk level into a preset pediatric fever course state machine template to instantiate and generate an individualized state machine that includes the acute fever phase, fever reduction observation period, recovery period, symptom recurrence pending judgment period, deterioration warning period, and event closure period, as well as their state transition conditions. Based on the preset duration and preset follow-up touchpoints of each state in the individualized state machine, a nursing content identifier and a self-assessment questionnaire identifier corresponding to each preset follow-up touchpoint are generated, and a post-diagnosis follow-up plan is formed according to the time sequence of each preset follow-up touchpoint and the corresponding nursing content identifier and the self-assessment questionnaire identifier.

4. The method according to claim 3, characterized in that, The calculation of the disease deviation based on the follow-up data, the medical history summary, and the diagnostic results includes: Extract body temperature time-series data, symptom quantitative scores, and new symptom entities from the follow-up data; extract historical body temperature peaks from the medical history summary; and match the corresponding expected recovery curve and expected symptom improvement trajectory for the same disease from the diagnostic results. Calculate the dynamic time curvature distance between the body temperature time series data and the expected recovery curve of the same disease, and normalize the dynamic time curvature distance to obtain the time series deviation. Calculate the absolute value of the residual between the symptom quantitative score and the expected symptom improvement trajectory at the same time point, and then normalize the weighted sum of the absolute values ​​of the residuals at each time point to obtain the symptom outcome deviation. The newly emerging symptom entity is matched with a pre-set list of danger signals. When the match is successful, the penalty value of the newly emerging danger signal is calculated according to the preset weight of the matched danger signal. When the match fails, the penalty value of the newly emerging danger signal is set to zero. The disease deviation is obtained by weighted summing of the time sequence deviation, the symptom outcome deviation, and the penalty value for newly emerging danger signals.

5. The method according to claim 4, characterized in that, The step of adjusting the post-diagnosis follow-up plan when the deviation of the condition reaches a preset trigger condition includes: The deviation degree of the condition is compared with a preset multi-level deviation threshold to obtain the deviation level; When the deviation level is mild, the post-diagnosis follow-up plan remains unchanged; When the deviation level is moderate, a temporary follow-up contact is inserted between two adjacent preset follow-up contacts in the post-diagnosis follow-up plan, and close observation guidance is pushed out at the temporary follow-up contact. When the deviation level is significant, the individualized state machine is forcibly migrated to the deterioration warning period. Based on the deterioration warning period, the preset follow-up touchpoints that are not currently executed in the post-diagnosis follow-up plan are suspended, and a high-frequency monitoring scheme is activated.

6. The method according to claim 5, characterized in that, The method further includes: After the high-frequency monitoring scheme is activated, the medical history summary, diagnosis results and reported follow-up data corresponding to the current fever event are retrieved, and the disease retrospective summary is generated by merging them according to the preset retrospective summary template. The disease retrospective summary is then pushed to the bound doctor's terminal.

7. The method according to claim 5, characterized in that, The generation of health records based on febrile events, using the multi-round dialogue information, triage results, diagnostic results, and follow-up data, includes: The multi-turn dialogue information, the triage results, the diagnosis results, and the follow-up data are written into the event log indexed by the fever event ID in chronological order to obtain the event log stream; The event log stream is aggregated by time dimension into a materialized view with the child ID as the primary key to obtain a health record. Each fever event record in the health record includes a symptom evolution map corresponding to the multi-turn dialogue information, an emergency change log corresponding to the triage result, a diagnosis conclusion corresponding to the diagnosis result, and a body temperature curve and outcome status corresponding to the follow-up data.

8. The method according to claim 1, characterized in that, The method further includes: Obtain the pre-diagnosis triage direction and the final diagnosis during diagnosis for the same fever event in the health record, perform structured alignment of the pre-diagnosis triage direction and the final diagnosis during diagnosis, and extract the difference features to obtain the diagnostic deviation set; Based on the diagnostic deviation set, a correction feedback package containing field-level modified records and diagnostic direction deviations is generated. The correction feedback package is then sent to the pre-diagnosis engine to update the pruning threshold and follow-up question trigger weight of the decision tree node.

9. A comprehensive management system for childhood fever, characterized in that, The system includes: The identification module is used to acquire multi-turn dialogue information related to fever in children, identify danger signals based on the multi-turn dialogue information, and generate triage results including medical history summary and triage level; The consultation module is used to synchronize the triage results to the doctor's end during the consultation and to obtain the diagnosis results input by the doctor based on the triage results; The follow-up module is used to generate a corresponding post-diagnosis follow-up plan based on the diagnosis results and the triage level. In the post-diagnosis stage, it obtains follow-up data reported based on the post-diagnosis follow-up plan and calculates the degree of deviation of the condition based on the follow-up data, the medical history summary and the diagnosis results. The early warning module is used to adjust the post-diagnosis follow-up plan and issue an early warning when the deviation of the condition reaches a preset trigger condition; The management module is used to generate health records based on the multi-round dialogue information, the triage results, the diagnosis results, and the follow-up data, with fever events as the unit.

10. A computer-readable storage medium storing a computer program executable on a processor, characterized in that, When the computer program is executed by the processor, it implements a method for managing the entire course of fever in children as described in any one of claims 1 to 8.

Images