Medical information-driven sepsis early warning method

Abstract

The invention relates to the technical field of intelligent medical diagnosis and treatment, in particular to a sepsis early warning method driven by medical information, which comprises the following steps of: acquiring combined physical sign data of a sepsis patient and screening frequency consistent sections, extracting slope change critical points to form a trend inflection point set, analyzing physical sign change directions before and after inflection points and extracting a synchronous direction change combination; and calculating a correlation coefficient to identify a pathology linkage relationship to obtain an early warning result. According to the method, the alignment precision of the multi-dimensional time series data is improved by synchronously collecting the multiple pieces of sign data and uniformly constructing the time series structure, the critical point position of sign slope change is extracted in the frequency consistency section, the trend change characteristics are collected, and further through trend consistency analysis of the sign change direction in the observation window, the alignment precision of the multi-dimensional time series data is improved. The capturing precision of the multi-sign cooperative variation event is improved, and frequency statistics and intensity analysis are introduced after sign combinations with consistent direction change are screened, so that the linkage relationship between abnormal signs is more stable and credible.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical diagnosis and treatment technology, and in particular to a medical information-driven method for early warning of sepsis. Background Technology

[0002] The field of intelligent medical diagnosis and treatment technology mainly involves combining traditional medical theories with modern information processing technology to achieve the collection, analysis and decision support of medical diagnostic information.

[0003] Among them, the medical information-driven sepsis early warning method refers to a means of early sepsis identification based on medical diagnosis and treatment information, which extracts features related to sepsis risk from subjective symptom information obtained during the medical diagnosis and treatment process.

[0004] Existing technologies rely solely on subjective symptom information extracted during medical diagnosis and treatment, lacking a time-series analysis mechanism for the frequency and trend direction of changes in multidimensional physical signs. This makes it prone to missed detections in the early stages of mild or atypical symptoms. In particular, when early signs of sepsis have not yet manifested as a single symptom, it is unable to effectively identify the potential linkage between multiple physical signs. It also lacks precise identification methods for the consistency of physical sign frequency and the sensitivity of slope changes, resulting in a single and unstable basis for judging early warning information. For example, when there are fluctuations in the patient's respiratory rate and tongue temperature but they have not reached the abnormal threshold, it is impossible to effectively combine these changes to judge the systemic pathological trend, thus limiting its ability to identify the latent risk of sepsis in its early stages. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a medical information-driven sepsis early warning method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a medical information-driven sepsis early warning method, comprising the following steps: S1: Synchronously collect combined signs data of sepsis patients during a specified period and screen for frequency consistency segments with consistent change frequencies in the combined signs data; S2: Identify the critical point where the slope change of the joint vital signs data in the frequency consistency segment reaches the preset change ratio, extract each critical point as a trend inflection point, and collect them into a trend inflection point set. S3: Set up observation windows before and after each trend inflection point in the set of trend inflection points, determine whether the direction and trend of change of each joint vital sign data in the window are the same, and extract the synchronous change of vital sign combination according to the density of the same inflection point; S4: Calculate the correlation coefficient of each sign in the synchronous directional sign combination, determine whether a pathological linkage relationship of sepsis is constituted based on the correlation coefficient, and obtain the sepsis early warning result.

[0007] The present invention is improved in that the frequency consistency segment includes a sign time index, a frequency alignment interval, and a joint sign label; the trend inflection point set includes the slope change direction, the change inflection point time, and the sign linkage mark; the synchronous change sign combination includes sign change pair, trend consistency index, and sign coupling mapping relationship; and the sepsis early warning result includes the pathological linkage pair number, the sign coordination intensity value, and the early warning time label.

[0008] The present invention is improved in that step S1 is specifically as follows: S101: Collect combined vital signs data of sepsis patients during a specified period. The combined vital signs data include heart rate, respiratory rate, body temperature, pulse rate and tongue temperature. Align each data item according to the sampling time to obtain combined vital signs data with a unified time structure. S102: Based on the unified time structure of the joint vital signs data, extract the peak and valley positions of each vital sign data in the continuous observation segment, calculate the interval length between adjacent extreme values ​​as the variation period index, and the fluctuation frequency of each vital sign at a specified time to obtain the joint vital signs data frequency sequence. S103: Based on the frequency sequence of the combined vital signs data, determine whether the frequency difference of each vital sign is lower than the frequency consistency judgment threshold in each time segment, number and mark the data segments that meet the frequency consistency condition, and extract the corresponding time index to obtain the frequency consistency segment.

[0009] The present invention is improved in that step S2 is specifically as follows: S201: Call the joint vital sign data under the corresponding time index in the frequency consistency segment, and extract the joint vital sign data trend sequence of the vital sign data changing with time according to the vital sign item; S202: Perform multi-segment linear fitting on the trend sequence of the combined vital signs data using the regression discontinuity method, calculate the slope of adjacent time points for each vital sign data after fitting, extract the change ratio between adjacent slopes in a continuous time period, and select the positions where the change ratio reaches the preset change ratio threshold as critical point time indexes to obtain the critical point index set. S203: Based on the time position recorded in the critical point index set, extract the corresponding vital signs and trend information, collect them according to the correspondence between vital signs and critical points, and establish a set of trend inflection points with trend change characteristics in the joint vital sign data.

[0010] The present invention is improved in that step S3 is specifically as follows: S301: Call the time index position of each trend inflection point in the trend inflection point set, set an observation window of equal length before and after each trend inflection point, extract the continuous numerical sequence of each vital sign item in the window, and obtain the trend inflection point observation data segment. S302: Based on the observed data segment of the trend inflection point, calculate the direction of change of the value of each joint vital sign data in the observation window, compare whether the direction of change of each vital sign in the same window is the same or whether the trend is consistent, extract the combination of vital sign items with the same direction of change and the corresponding inflection point index, and obtain the preliminary combination pair of synchronous change of vital signs. S303: Based on the preliminary combination of synchronous directional signs, the frequency and distribution of each combination of signs in the trend inflection point set are statistically analyzed, the degree of dense distribution in the continuous inflection point sequence is calculated, and the combination of signs with an inflection point density higher than the set standard is selected as a stable linkage structure to obtain the synchronous directional sign combination.

[0011] The present invention is improved in that step S4 is specifically as follows: S401: Call the combination of vital signs in the synchronous directional vital signs combination, extract the original numerical sequence according to the trend inflection point time segment corresponding to each combination of vital signs, and obtain the synchronous directional joint vital signs sequence set. S402: Based on the synchronous change-direction joint sign sequence set, calculate the Pearson correlation coefficient between the observation sequences of two signs in each group of sign items, and screen the sign combinations with correlation coefficients lower than the preset negative correlation threshold and corresponding signs with opposite change directions to obtain pathological linkage sign pairs. S403: Based on the combination of signs and corresponding trend inflection points recorded in the pathological linkage sign pairing, extract all the signs linkage information that constitute the pathological linkage relationship in the current time period to obtain the sepsis early warning result.

[0012] The present invention is improved and further includes S5: performing structural classification of the linkage relationship in the sepsis early warning result, classifying it into the corresponding risk response category, and obtaining the risk classification result; The risk classification results specifically include risk level identifier, linkage mode category, and vital sign structure type.

[0013] The present invention is improved in that step S5 is specifically as follows: S501: Call the combined sign data marked as pathological linkage in the sepsis early warning result, extract the quantity, location distribution and direction of coordinated change of each group of pathological linkage sign pairs within the trend inflection point time period, and obtain the set of linkage performance characteristics. S502: Based on the aforementioned set of linkage performance characteristics, classify and group the pathological linkage signs according to their quantity, trend inflection point concentration, and directional consistency. Call the risk response pattern category defined in the structural classification standard, compare the classification results with the structural characteristics of the established patterns, determine the risk structure label, and obtain the risk category attribution list. S503: Based on the labels listed in the risk category attribution list, label each linkage relationship in the sepsis early warning result with the corresponding risk response category to obtain the risk classification result.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by synchronously collecting multiple vital sign data and uniformly constructing a time series structure, the alignment accuracy of multidimensional time series data is improved. The critical point positions of vital sign slope changes are extracted in frequency consistency segments, and trend change characteristics are collected, enhancing the ability to identify abnormal vital sign dynamics. Furthermore, by analyzing the trend consistency of vital sign change directions within the observation window, the accuracy of capturing multi-vital sign synergistic variation events is improved. After screening for combinations of vital signs with consistent changes, frequency statistics and density analysis are introduced, making the linkage relationship between abnormal vital signs more stable and reliable. Finally, by combining the negative correlation and directional reversal between vital signs, pathological linkage structures are identified, and risk classification is completed. This makes the identification of potential pathological mechanisms of sepsis more systematic, and the judgment criteria for early warning results more accurate. Overall, a multi-level judgment system from dynamic signal frequency, trend direction, linkage coupling to risk attribution is realized, enhancing the timeliness and reliability of early identification of sepsis risk. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method proposed in this invention; Figure 2 This is a detailed flowchart of step S1 of the present invention; Figure 3 This is a detailed flowchart of step S2 of the present invention; Figure 4 This is a detailed flowchart of step S3 of the present invention; Figure 5 This is a detailed flowchart of step S4 of the present invention; Figure 6 This is a detailed flowchart of step S5 of the present invention. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0017] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0018] Please see Figure 1 This invention provides a technical solution: a medical information-driven method for sepsis early warning, comprising the following steps: S1: Synchronously collect combined signs data of sepsis patients during a specified period and screen for frequency consistency segments with consistent change frequencies in the combined signs data; S2: Identify the critical point where the slope of the joint trait data in the frequency consistency segment reaches the preset change ratio, extract each critical point as a trend inflection point, and collect them into a trend inflection point set. S3: Set up observation windows before and after each trend inflection point in the trend inflection point set, determine whether the direction and trend of change of each joint vital sign data in the window are the same, and extract synchronous change of vital sign combinations according to the density of the same inflection point. S4: Calculate the correlation coefficient of each sign in the synchronous change sign combination, determine whether a pathological linkage relationship of sepsis is constituted based on the correlation coefficient, and obtain the sepsis early warning result; The frequency consistency segment includes the sign time index, frequency alignment interval, and joint sign label; the trend inflection point set includes the slope change direction, change inflection point time, and sign linkage marker; the synchronous change sign combination includes sign change pair, trend consistency index, and sign coupling mapping relationship; and the sepsis early warning result includes the pathological linkage pair number, sign coordination strength value, and early warning time label.

[0019] Please see Figure 2 Step S1 is as follows: S101: Collect combined vital signs data of sepsis patients during a specified period. The combined vital signs data include heart rate, respiratory rate, body temperature, pulse rate and tongue temperature. Align each data item according to the sampling time to obtain combined vital signs data with a unified time structure. First, the sampling population should be identified as confirmed sepsis patients, and the sampling period should be clearly defined, such as three consecutive days from the time of admission. Each day, three fixed time points—morning, noon, and evening—sampling nodes should be selected. At each node, five vital signs data should be collected: heart rate, respiratory rate, body temperature, pulse rate, and tongue temperature. Each sampling should be timestamped. Structured records should be exported from the ward monitoring equipment to generate a data table containing time columns and columns for each vital sign. For any gaps in the time points of different vital sign data, interpolation or previous values ​​should be used to fill in the gaps to construct a unified time structure. For example, if there are gaps in the heart rate data at 08:00, ... For body temperature and tongue temperature data, but missing respiratory rate and pulse rate data, the corresponding data at 07:00 and 09:00 are searched. The interpolation value at 08:00 is calculated by dividing (the value at 09:00 + the value at 07:00) by 2, or the value of the vital sign at 07:00 is used as the filler value. After unification, each row in the data table is a unified time point, and each column contains all five vital signs, in the form of Time: 08:00, Heart Rate: Value, Respiratory Rate: Value, Body Temperature: Value, Pulse Rate: Value, Tongue Temperature: Value. The data at each sampling point on the entire time axis is complete, forming a unified time structure joint vital sign data table.

[0020] S102: Based on the unified time structure of joint vital signs data, extract the peak and valley positions of each vital sign data in the continuous observation segment, calculate the interval length between adjacent extreme values ​​as the variation period index, and the fluctuation frequency of each vital sign at a specified time to obtain the joint vital signs data frequency sequence. Based on unified time-structured joint vital sign data, a sliding analysis window is set for each vital sign, such as a six-hour window. The interval between the occurrence of the maximum and minimum values ​​within the window is calculated and defined as the variation period. For example, if the maximum heart rate is observed to occur at the 90th minute and the minimum at the 30th minute within a six-hour window, then the variation period = 90 minutes - 30 minutes = 60 minutes. The variation periods of respiratory rate, body temperature, pulse rate, and tongue temperature are calculated accordingly. Then, the number of times the extreme values ​​(maximum and minimum) of the vital sign alternate within the same time period is counted. If it occurs 4 times within six hours, then the fluctuation frequency = 4 times ÷ 6 hours = 0.67 times per hour. This frequency value is the frequency expression value of the vital sign in that window. By sliding the time window segment by segment, a complete frequency sequence of the vital sign is formed. The frequency sequences of the five vital signs are constructed in this way. Finally, the frequency values ​​of heart rate, respiratory rate, body temperature, pulse rate, and tongue temperature in each time period are combined to form a multidimensional time series, which constitutes the joint vital sign data frequency sequence.

[0021] S103: Based on the frequency sequence of the combined vital signs data, determine whether the frequency difference of each vital sign in each time segment is lower than the frequency consistency judgment threshold, number and mark the data segments that meet the frequency consistency condition and extract the corresponding time index to obtain the frequency consistency segment. Based on the frequency sequence of combined vital signs data, within each analysis window, the frequency values ​​of the five vital signs are compared to see if they are within the same range. The determination method is to see if the difference between the frequencies of any two vital signs is lower than a preset threshold. For example, if the frequency consistency threshold is set to 0.03 times per hour, then the frequency difference between any two vital signs must meet the following form: if the heart rate is 0.52 times per hour and the respiratory rate is 0.50 times per hour, the difference is calculated as 0.52 - 0.50 = 0.02 times per hour, which is less than the threshold of 0.03, and the two are considered to be consistent. This process is repeated for each pair of combinations (a total of ten pairs) for comparison. As long as the difference between each pair of frequency differences is less than 0.03 times per hour, the time period is considered to have frequency consistency. The start and end times are recorded as consistent segments. For example, if the segment starts at the 3rd hour and ends at the 9th hour, then the segment is defined as frequency consistency segment 1, numbered FZ1. This process is repeated for all time periods, extracting all time segments that meet the conditions, and then numbering and outputting them as a set of frequency consistency segments.

[0022] Please see Figure 3 Step S2 is as follows: S201: Call the joint vital signs data under the corresponding time index in the frequency consistency segment, and extract the joint vital signs data trend sequence of changes over time according to the vital signs item; To retrieve joint vital sign data under the corresponding time index within a frequency consistency segment, the start and end time indices of each frequency consistency segment must first be determined. For example, for frequency segment 1, which runs from the 2nd to the 8th hour, all record rows within that segment are extracted from the time column of the corresponding data table to form a subset of the five vital signs for that time period. Within this subset, the value of each vital sign is extracted individually according to the time series to construct a vital sign trend sequence. For instance, starting from the 2nd hour, the heart rate values ​​are extracted and arranged into a continuous sequence at a sampling frequency of once every 5 minutes, thus forming the heart rate trend sequence. Similarly, the trend sequences of respiratory rate, body temperature, pulse rate, and tongue temperature are extracted. Each trend sequence corresponds to the original data time axis. For example, if a frequency consistency segment covers 72 data records, it means that the trend sequence has vital sign values ​​at 72 time points, forming a time series array. The position of each vital sign data in the sequence matches the sampling time on the time axis. After performing this process on all frequency consistency segments, a set of joint vital sign trend data containing continuous time and complete vital signs is constructed.

[0023] S202: The trend sequence of joint vital signs data is linearly fitted in multiple segments by the regression discontinuity method. The slope of each vital sign data point adjacent to the time point after fitting is calculated. The change ratio between adjacent slopes in the continuous time period is extracted. The positions where the change ratio reaches the preset change ratio threshold are selected as critical point time indices to obtain the critical point index set. The time trend sequence of each vital sign (such as body temperature, heart rate, etc.) is extracted based on frequency consistency segments. Structurally, this sequence may not be a single linear increase or decrease, but rather contains multiple segments with different rates of change. To identify these rate abrupt changes, a multi-discontinuity regression model is applied to the trend sequence of each vital sign. Through piecewise fitting, the changes in slope between segments are extracted, and these changes are used to determine whether a significant trend reversal has occurred, thereby determining the "critical point index set".

[0024] The regression discontinuity modeling form is as follows: ; in, This represents the predicted fitted value of a specified vital sign (such as body temperature) at a given time point. For example, the observed value of body temperature is the response variable of the model. This represents the sampling time, which can be in minutes or as a location number, and serves as an explanatory variable for the model. It is the intercept of the regression start segment, representing the baseline value of vital signs at the start time; It is the slope of the first linear segment, representing the magnitude of change of vital signs per unit time in the first segment; It is the first Each breakpoint indicates a location where the trend of vital signs changes; It is the first The slope change value added after each breakpoint; It is an indicator function; a value of 1 indicates that the breakpoint has been exceeded at the current time point. Otherwise, it is 0; This is the error term, representing the residual deviation between the model's predicted values ​​and the actual observed values. It is defined as follows: Error Term Based on the premise of "zero average error and limited fluctuation," this means that all fitting residuals should be statistically approximately symmetrically distributed around zero. The fluctuation range is not predetermined to a specific value, but rather is based on the difference between the observed vital signs data and the model predictions during the actual fitting process. During regression fitting, the system automatically estimates and adjusts the overall error fluctuation to ensure that it does not systematically interfere with the direction of trend changes. The number of breakpoints refers to the total number of breakpoints preset or estimated in the model, representing the rate of change of vital signs in the trend sequence. Substructural change.

[0025] After fitting, the slope values ​​of each vital sign over time can be clearly obtained based on the linear expression of each segment in the model. These slopes reflect the rate of increase or decrease of the vital sign within each short time interval. The slopes of these continuous segments are extracted to form a slope sequence, denoted as . .

[0026] Furthermore, for each pair of adjacent slopes and Calculate its change ratio, defined as: ; The change ratio This reflects the relative magnitude of the rate of change in vital signs between adjacent time periods. To identify locations of abrupt trend changes, a threshold value for the change ratio needs to be set, denoted as . For each change ratio, filter the values ​​if any of the following conditions are met: This indicates a rapid increase in the rate of change; This indicates a significant slowdown in the rate of change; therefore, the time point corresponding to this position... As a critical point of a trend, that is, one of the turning points of a trend.

[0027] Finally, all location time points that meet the threshold filtering criteria are... The data is collected to form a critical point index set for each vital sign. Each time index in this set identifies a structural inflection point in the trend curve of the vital sign, which is used for subsequent aggregation of trend inflection point sets and trend classification processing. The entire processing flow is performed independently for each of the five vital signs, ensuring that the critical point of each vital sign originates only from its own trend behavior and does not cause cross-interference.

[0028] Suppose that body temperature data were collected at 6 consecutive time points (sampled every 10 minutes) within a certain frequency consistency range, and the time points are as follows: minute, minute, minute, minute, minute, minutes; the corresponding body temperature reading is: ℃, ℃, ℃, ℃, ℃, ℃.

[0029] Based on observation, let's assume we set two breakpoints: Breakpoint 1: Minutes (3rd time point), Breakpoint 2: Minutes (5th time point); The body temperature trend sequence was divided into 3 segments for linear fitting: Section 1: arrive Minutes, corresponding points ; Section 2: arrive Minutes, corresponding points ; Section 3: arrive Minutes, corresponding points .

[0030] Fit a linear function to each segment The slope of each segment can be calculated separately. : Section 1: The body temperature rose from 36.5℃ to 37.4℃, expressed by a slope of 1= ℃ / minute; Section 2: The body temperature rose from 37.4℃ to 37.2℃, using a slope of 2= ℃ / minute; Section 3: The body temperature dropped from 37.2℃ to 36.9℃, using a slope of 3= ℃ / minute; Therefore, the slope sequence of the body temperature trend in the three time periods is obtained as follows: , , .

[0031] Change ratio 1 (segment 1 and segment 2): ; Change ratio 2 (segment 2 and segment 3): .

[0032] Set the threshold for the change ratio right Not satisfied However, the sign changes abruptly, from positive to negative ⇒ the trend direction changes, which can be judged as a trend abrupt change ⇒ mark breakpoint 1 (20 minutes) as the critical point; for :satisfy ⇒The trend is accelerating downwards⇒Mark breakpoint 2 (40 minutes) as the critical point; the critical point index set ultimately contains two trend inflection points: minute, Minutes. This index set indicates significant abrupt changes in the rate of change of body temperature at time points 3 and 5, serving as the source input for subsequent inflection point sets. This process can also be performed independently on other vital signs, extracting the corresponding critical point time indexes using trend analysis methods.

[0033] First, using piecewise linear regression, the continuous data on vital signs changing over time is divided into multiple time periods. For each period, the rate of change, or slope, is calculated, indicating whether the vital sign is rising, falling, or remaining stable within that time period. Then, the slopes of these consecutive periods are compared to calculate the ratio between adjacent slopes, reflecting the magnitude of the difference in the rate of change of the vital sign across different time periods. This ratio indicates whether the trend change is significant. When a certain ratio reaches a preset threshold, a sudden change in the vital sign is considered to have occurred at that time point, indicating a shift from slow change to rapid rise or fall, or a reversal of the direction of change. Finally, through this series of steps—fitting, calculating slopes, and comparing slope ratios with thresholds—critical time points in the trend of vital sign changes are identified, establishing a time index set for trend abrupt changes. This provides a foundation for subsequent trend feature classification and trend inflection point structure extraction. The entire process is characterized by its clear structure, repeatability, and quantifiable indicators.

[0034] S203: Based on the time position recorded in the critical point index set, extract the corresponding vital signs and trend information, collect them according to the correspondence between vital signs and critical points, and establish a set of trend inflection points with trend change characteristics in the joint vital signs data. Based on the constructed critical point index set, the original vital sign trend sequence is first retrieved at each time position in the index set. The vital sign name and trend information corresponding to that time point are extracted, namely the type of vital sign (such as body temperature, heart rate, etc.) at that time point and the direction of change of the fitted segment, such as from rising to falling, or from falling to accelerating rising. Then, according to the position of the time point in the trend, it is classified as the trend inflection point under that vital sign item, while retaining the time position and the nature of the trend change. On this basis, all extracted trend inflection points are classified by vital sign item. Multiple inflection points belonging to the same vital sign item are grouped together and formed into an inflection point list with time position and trend change description. After completing this collection process for all five vital signs, a set of trend inflection points in the joint vital sign data is constructed as a whole. Each inflection point record in this set contains the vital sign name, the time point of occurrence, the direction of trend reversal, etc.

[0035] Please see Figure 4 Step S3 is as follows: S301: Call the time index position of each trend inflection point in the trend inflection point set, set an observation window of equal length before and after each trend inflection point, extract the continuous numerical sequence of each vital sign item in the window, and obtain the trend inflection point observation data segment. In the trend inflection point set, each inflection point contains its corresponding time index and structural features of vital sign changes. Next, observation windows of equal length need to be set before and after each inflection point. For example, if the inflection point time is a certain minute, then five minutes can be set before and after that minute as the window range. Within this observation window, continuous recorded values ​​of five vital signs—body temperature, heart rate, tongue temperature, respiratory rate, and pulse rate—are extracted from the combined vital sign data, ensuring that each vital sign has complete time point coverage within the window and maintaining time alignment. Taking respiratory rate as an example, continuous measurements of this vital sign are extracted from the beginning to the end of the window, forming a local change sequence of this vital sign near the current inflection point; the other vital signs are processed in the same way. Finally, for each inflection point, a set of continuous time values ​​of the five vital signs is formed, called the trend inflection point observation data segment.

[0036] S302: Based on the trend inflection point observation data segment, calculate the direction of change of each joint vital sign data in the observation window, compare whether the direction of change of each vital sign in the same window is the same or whether the trend is consistent, extract the combination of vital sign items with the same direction of change and the corresponding inflection point index, and obtain the preliminary combination pair of synchronous change of vital signs. Based on the observation of data segments at trend inflection points, it is necessary to analyze the direction of change of each of the five vital signs within the observation window. The method is as follows: take the vital sign value at the last time point of the latter half of the window and subtract the vital sign value at the first time point of the first half of the window; this calculates the overall change of the vital sign within the observation interval. For example, to determine respiratory rate, the calculation is: End-tidal respiratory rate - Initial respiratory rate = Change in respiratory rate; a positive result indicates increased respiration (rising direction); a negative result indicates decreased respiration (falling direction). Similarly, for body temperature, the calculation is: End-tidal body temperature - Initial body temperature = Direction of body temperature change; for pulse rate, the calculation is: End-tidal pulse rate - Initial pulse rate = Direction of pulse rate change; heart rate and tongue temperature are calculated in the same way. After determining the direction of all vital signs, compare whether the changing directions of the five vital signs are consistent. If three or more vital signs show the same changing direction, such as all three vital signs being upward, then these three vital signs are grouped into a synchronous changing direction combination. This combination is then paired with its corresponding trend inflection point time index to form a preliminary structure, which is called the preliminary synchronous changing direction vital sign combination pair.

[0037] S303: Based on the initial combination of synchronous directional vital signs, the frequency and distribution of each combination of vital signs in the trend inflection point set are statistically analyzed, the degree of dense distribution in the continuous inflection point sequence is calculated, and vital sign combinations with inflection point density higher than the set standard are selected as stable linkage structures to obtain synchronous directional vital sign combinations. After summarizing all preliminary combinations of synchronously changing vital signs, statistical analysis is performed on these combinations. First, the frequency of each combination is counted within the entire set of trend inflection points; this frequency is called the combination's occurrence frequency. For example, if the combination of body temperature, pulse rate, and respiratory rate appears consecutively within multiple inflection point windows, its frequency is its total occurrence count. Second, the temporal distribution density of these combinations is assessed, i.e., whether these combinations are concentrated in a certain time period in the time series. This is done by calculating the sum of the time intervals between adjacent inflection points. For example, if a combination appears consecutively in multiple inflection points with intervals not exceeding ten minutes, the distribution density is high. The judgment criterion is: the last inflection point occurrence time minus the initial inflection point occurrence time = time span. Dividing this span by the number of times the combination appears yields the average occurrence density. If this density value is less than a set upper limit, it indicates that the combination structure is relatively compact. Finally, all combinations were screened using both density and frequency. The combination of signs that simultaneously exhibited high frequency of occurrence and high density of distribution in the trend inflection point set was selected and defined as the synchronous change-direction sign combination, representing a stable sign structure with significant linkage and directional consistency.

[0038] Please see Figure 5 Step S4 is as follows: S401: Call the combination of vital signs in the synchronous directional vital signs combination, extract the original numerical sequence according to the trend inflection point time segment corresponding to each group of vital signs combination, and obtain the synchronous directional joint vital signs sequence set. After calling the identified vital sign combinations from the synchronized directional vital sign combinations, the trend inflection point time point corresponding to each vital sign combination needs to be used as a reference. The continuous numerical sequence of each item of the vital sign combination before and after the trend inflection point time point needs to be extracted from the original data of the combined vital signs according to the time axis. For example, if a combination includes three vital signs: body temperature, respiratory rate, and heart rate, a fixed time window is set at the corresponding inflection point, such as five minutes before and after. The continuous numerical arrangement of body temperature, respiratory rate, and heart rate within the window is extracted from the original combined vital sign data to construct the combined time series record of this set of vital signs under this time period. This operation process is repeated, and the corresponding original sequence extraction processing is performed on each synchronized directional vital sign combination and its associated inflection point. Finally, the results are summarized to form a synchronized directional combined vital sign sequence set. Each sequence retains its corresponding vital sign combination information and trend inflection point index.

[0039] S402: Based on the synchronous direction-changing joint sign sequence set, calculate the Pearson correlation coefficient between the observation sequences of two signs in each group of sign items, and screen the sign combinations with correlation coefficients lower than the preset negative correlation threshold and corresponding signs with opposite directions of change to obtain pathological linkage sign pairs. Based on a set of synchronized directional joint vital sign sequences, to determine whether there is a reverse-trend linkage within each vital sign combination, the correlation between the changing behaviors of any two vital sign items in each combination is calculated, and the Pearson correlation coefficient method is used to measure the degree of linear correlation. This method is suitable for assessing the synchronicity and directionality between two time series of equal length, and is particularly useful for identifying vital sign pairs that still exhibit significant synchronous fluctuations despite opposite trends.

[0040] The formula for calculating the Pearson correlation coefficient is: ; in, : Represents the Pearson correlation coefficient between vital sign A and vital sign B. It is an indicator that measures whether the numerical fluctuations of these two vital signs have a linear correlation within a time window, and its value ranges from -1 to 1. : Represents the original observation value of vital sign A at the k-th time point, such as the body temperature value at the k-th minute in the observation window; : Represents the raw observation value of vital sign B at the same k-th time point, such as the heart rate measurement at that minute; : Represents the average value of vital sign A over the entire observation window. It is calculated by summing the values ​​of vital sign A at all time points and dividing by the total number of time points. : This represents the average value of sign B over the entire observation window, and the processing method is the same as that for sign A; : This is the index of the time point in the window, used to indicate which sampling point in the time series, ranging from 1 to N; : This represents the total number of time points included in the observation window. For example, if the sampling occurs once per minute in a five-minute window before and after the observation window, then N is 11.

[0041] Suppose there are 5 time points in the observation window at a certain trend inflection point, and the following vital sign data were collected: Body temperature sequence (A): 36.6, 36.7, 36.8, 36.9, 37.0; Heart rate sequence (B): 94, 92, 90, 88, 86.

[0042] Step 1: Calculate the mean: ; ; Step 2: Calculate the numerator: ; Step 3: Calculate the denominator: Sum of squares of body temperature deviation: ; Sum of squares of heart rate deviation: ; Square root part: , ; Step 4: Substitute into the formula to calculate the correlation coefficient: .

[0043] The results showed that the observed body temperature gradually increased, while the observed heart rate continuously decreased, indicating that the two vital signs changed in opposite directions over time. The Pearson correlation coefficient was negative one, indicating that they not only changed in completely opposite directions but also had a highly linear inverse relationship. Based on the judgment criteria, if two vital signs change in opposite directions within the same time window and their correlation coefficient is below a set negative correlation threshold (e.g., -0.5), they can be considered to have significant pathological linkage characteristics. Therefore, in this example, this pair of vital signs simultaneously meets the two judgment conditions of opposite direction and high negative correlation, conforming to the criteria for identifying a pathological linkage pair of vital signs.

[0044] The threshold is a pre-defined boundary value based on domain experience or statistical distribution, used to distinguish between weak and significant negative correlations. Pairs of signs that meet the criteria will be recorded as pathologically linked sign pairs, while retaining their corresponding trend inflection point time information, forming a key data combination with an inverse relationship, used for subsequent sepsis risk structure identification and early warning signal extraction. Each pathologically linked sign pair undergoes dual verification through trend direction judgment and quantitative correlation calculation to ensure it has structural linkage evidence at the data level.

[0045] S403: Based on the combination of signs and corresponding trend inflection points recorded in the pathological linkage sign alignment, extract all signs linkage information that constitute the pathological linkage relationship in the current time period to obtain sepsis early warning results. Based on the combination of signs and their corresponding trend inflection points recorded in the pathological linkage sign pairs, it is necessary to uniformly extract and collect all signs with linkage relationships in the current time period. First, all marked pathological linkage sign pairs are traversed, and they are located in the original data of joint signs according to their corresponding trend inflection point time index. All sign pairs in the same time period are combined and classified. For example, if multiple sign pairs satisfy the pathological linkage relationship at a certain trend inflection point time, these sign pairs are combined to form the sign linkage structure at the current time point. At the same time, the total number of linked signs included at the time point, the change direction attribute of each linkage pair, and related feature values ​​are marked. On this basis, it is determined whether the sepsis warning trigger condition is met. For example, if the pathological linkage structure continues to appear in a continuous time period, or the number of linkage pairs in the same time point exceeds the preset standard, it is determined that there is an abnormal coordinated change state in the current time period, and the sign linkage details of the time period are output as the sepsis warning result.

[0046] Please see Figure 6 It also includes S5: Structurally classifying the linkage relationships in the sepsis early warning results and assigning them to the corresponding risk response categories to obtain risk classification results; The risk classification results are specifically defined as risk level identifier, linkage mode category, and vital sign structure type. Step S5 is as follows: S501: Call the combined signs data marked as pathological linkages in the sepsis early warning results, extract the quantity, location distribution and direction of coordinated change of each group of pathological linkage signs within the trend inflection point time period, and obtain the set of linkage performance characteristics. After retrieving the combined signs data marked as pathologically linked from the sepsis early warning results, it is necessary to extract the linkage performance of each pair of pathologically linked signs within the trend inflection point time period, including the number, distribution location, and direction of the sign pairs. First, the total number of signs identified as pathologically linked pairs within each trend inflection point time period is counted as one of the linkage strength indicators. Second, the interval between the time point of these sign pairs and the trend inflection point time point is recorded. For example, if a sign pair occurs two minutes before the inflection point, the calculation is: trend inflection point time minus the occurrence time of the linked pair = occurrence location offset. Finally, the direction of change of the sign pairs is categorized. For example, if sign A increases and sign B decreases, the direction relationship is opposite, indicated by the final value of one sign minus the initial value of the other sign being positive, and the final value of the other sign minus the initial value being negative, indicating opposite directions. If both are increasing or both are decreasing, the directions are consistent, indicating linkage in the same direction. These linkage quantities, distribution locations, and direction attributes are combined and recorded to constitute the linkage performance characteristic set for that time period.

[0047] S502: Based on the set of linkage performance characteristics, the pathological linkage signs are classified and grouped according to the quantity, trend inflection point concentration and directional consistency of the pairs. The risk response pattern categories defined in the structural classification standard are called, and the degree of consistency between the classification results and the structural characteristics of the established patterns is compared to determine the risk structure label and obtain the risk category attribution list. Based on the set of linked performance characteristics, all pathological linked sign pairs are categorized. The categorization criteria include: First, the number of sign pairs appearing within the same time period, serving as an indicator of the scale of the linkage, calculated as the sum of the number of linked pairs within the same time period; Second, the concentration of distance between trend inflection point time points, called the concentration degree, calculated as the maximum value minus the minimum value of the trend inflection point = time distribution span. If this span is less than a preset length, the inflection points are considered highly concentrated; Third, the degree of directional consistency, i.e., the proportion of the direction type with the largest proportion among all sign pairs, calculated as the number of linkages in the same direction divided by the total number of linkages = directional consistency rate. After calculating these three indicators, each group of linkage structures is divided into different feature types according to the above three dimensions, and a risk response pattern template pre-defined in the structure classification standard is used for matching. For example, if the standard template defines high concentration, high consistency, and large scale as high-risk linkage patterns, then linkage structures that meet this combination of characteristics are compared as high-risk categories. After all comparisons are completed, the risk label of each group of linked structures is determined, a risk category attribution list is formed, and the type number, label level and attribution basis of each group of structures are recorded.

[0048] S503: Based on the labels listed in the risk category attribution list, label each linkage in the sepsis early warning result with the corresponding risk response category to obtain the risk classification result; Based on the risk labels identified in the risk category attribution list, each group of pathological linkage signs in the sepsis early warning results is labeled. First, the time index and sign combination information corresponding to each linkage structure are read. Then, its matching label in the risk attribution list is searched, and the label content is written into the attribute identifier of that linkage structure. The labeled content includes risk level, response mode type, and structure classification number. For example, if a structure belongs to a high-density, high-intensity linkage group and has strong directional consistency, it is labeled as a high-risk category; if another structure has scattered sign distribution and inconsistent directional fluctuations, it is classified as a low-risk category. The entire labeling process is completed by matching structures to their corresponding labels, ensuring that each pathological linkage structure has a clear risk attribution attribute. Finally, a complete set of risk classification results is formed, indicating the response level and type of all linkage structures under the current sepsis early warning system.

[0049] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A medical information-driven method for sepsis early warning, characterized in that, Includes the following steps: S1: Synchronously collect combined signs data of sepsis patients during a specified period and screen for frequency consistency segments with consistent change frequencies in the combined signs data; S2: Identify the critical point where the slope change of the joint vital signs data in the frequency consistency segment reaches the preset change ratio, extract each critical point as a trend inflection point, and collect them into a trend inflection point set. S3: Set up observation windows before and after each trend inflection point in the set of trend inflection points, determine whether the direction and trend of change of each joint vital sign data in the window are the same, and extract the synchronous change of vital sign combination according to the density of the same inflection point; S4: Calculate the correlation coefficient of each sign in the synchronous directional sign combination, determine whether a pathological linkage relationship of sepsis is constituted based on the correlation coefficient, and obtain the sepsis early warning result.

2. The medical information-driven sepsis early warning method according to claim 1, characterized in that: The frequency consistency segment includes a sign time index, a frequency alignment interval, and a joint sign label. The trend inflection point set includes the slope change direction, the change inflection point time, and the sign linkage marker. The synchronous change sign combination includes sign change pairs, trend consistency indicators, and sign coupling mapping relationships. The sepsis early warning result includes the pathological linkage pair number, the sign coordination intensity value, and the early warning time label.

3. The medical information-driven sepsis early warning method according to claim 1, characterized in that: Step S1 is as follows: S101: Collect combined vital signs data of sepsis patients during a specified period. The combined vital signs data include heart rate, respiratory rate, body temperature, pulse rate and tongue temperature. Align each data item according to the sampling time to obtain combined vital signs data with a unified time structure. S102: Based on the unified time structure of the joint vital signs data, extract the peak and valley positions of each vital sign data in the continuous observation segment, calculate the interval length between adjacent extreme values ​​as the variation period index, and the fluctuation frequency of each vital sign at a specified time to obtain the joint vital signs data frequency sequence. S103: Based on the frequency sequence of the combined vital signs data, determine whether the frequency difference of each vital sign is lower than the frequency consistency judgment threshold in each time segment, number and mark the data segments that meet the frequency consistency condition, and extract the corresponding time index to obtain the frequency consistency segment.

4. The medical information-driven sepsis early warning method according to claim 1, characterized in that: Step S2 is as follows: S201: Call the joint vital sign data under the corresponding time index in the frequency consistency segment, and extract the joint vital sign data trend sequence of the vital sign data changing with time according to the vital sign item; S202: Perform multi-segment linear fitting on the trend sequence of the combined vital signs data using the regression discontinuity method, calculate the slope of adjacent time points for each vital sign data after fitting, extract the change ratio between adjacent slopes in a continuous time period, and select the positions where the change ratio reaches the preset change ratio threshold as critical point time indexes to obtain the critical point index set. S203: Based on the time position recorded in the critical point index set, extract the corresponding vital signs and trend information, collect them according to the correspondence between vital signs and critical points, and establish a set of trend inflection points with trend change characteristics in the joint vital sign data.

5. The medical information-driven sepsis early warning method according to claim 4, characterized in that: The trend sequence of the combined vital signs data is subjected to multi-segment linear fitting using the following formula: ； Calculate the predicted fitted values ​​of a specified vital sign at a given time point. ; in, This indicates the sampling time, which can be in minutes or a location number. It is the intercept of the regression start segment, representing the baseline value of vital signs at the start time. It is the slope of the first linear segment, representing the magnitude of change of vital signs per unit time in the first segment. It is the first Each breakpoint represents a location where the trend of vital signs changes. It is the first The slope change value added after each breakpoint, It is an indicator function; a value of 1 indicates that the breakpoint has been exceeded at the current time point. Otherwise, it is 0. This is the error term, representing the residual deviation between the predicted value and the actual observed value. It represents the number of breakpoints.

6. The medical information-driven sepsis early warning method according to claim 1, characterized in that: Step S3 is as follows: S301: Call the time index position of each trend inflection point in the trend inflection point set, set an observation window of equal length before and after each trend inflection point, extract the continuous numerical sequence of each vital sign item in the window, and obtain the trend inflection point observation data segment. S302: Based on the observed data segment of the trend inflection point, calculate the direction of change of the value of each joint vital sign data in the observation window, compare whether the direction of change of each vital sign in the same window is the same or whether the trend is consistent, extract the combination of vital sign items with the same direction of change and the corresponding inflection point index, and obtain the preliminary combination pair of synchronous change of vital signs. S303: Based on the preliminary combination of synchronous directional signs, the frequency and distribution of each combination of signs in the trend inflection point set are statistically analyzed, the degree of dense distribution in the continuous inflection point sequence is calculated, and the combination of signs with an inflection point density higher than the set standard is selected as a stable linkage structure to obtain the synchronous directional sign combination.

7. The medical information-driven sepsis early warning method according to claim 1, characterized in that: Step S4 is as follows: S401: Call the combination of vital signs in the synchronous directional vital signs combination, extract the original numerical sequence according to the trend inflection point time segment corresponding to each combination of vital signs, and obtain the synchronous directional joint vital signs sequence set. S402: Based on the synchronous change-direction joint sign sequence set, calculate the Pearson correlation coefficient between the observation sequences of two signs in each group of sign items, and screen the sign combinations with correlation coefficients lower than the preset negative correlation threshold and corresponding signs with opposite change directions to obtain pathological linkage sign pairs. S403: Based on the combination of signs and corresponding trend inflection points recorded in the pathological linkage sign pairing, extract all the signs linkage information that constitute the pathological linkage relationship in the current time period to obtain the sepsis early warning result.

8. The medical information-driven sepsis early warning method according to claim 7, characterized in that: To calculate the Pearson correlation coefficient between the observation sequences of two vital signs in each combination of vital signs, the formula is: ； in, : Represents the Pearson correlation coefficient between symptom A and symptom B. This represents the original observation value of symptom A at the k-th time point. This represents the original observation value of trait B at the same k-th time point. This represents the average value of symptom A over the entire observation window. This represents the average value of symptom B over the entire observation window. For the time point index in the window, This represents the total number of time points contained in the observation window.

9. The medical information-driven sepsis early warning method according to claim 1, characterized in that: It also includes S5: Structurally classifying the linkage relationships in the sepsis early warning results and assigning them to the corresponding risk response categories to obtain risk classification results; The risk classification results specifically include risk level identifier, linkage mode category, and vital sign structure type.

10. The medical information-driven sepsis early warning method according to claim 9, characterized in that: Step S5 is as follows: S501: Call the combined sign data marked as pathological linkage in the sepsis early warning result, extract the quantity, location distribution and direction of coordinated change of each group of pathological linkage sign pairs within the trend inflection point time period, and obtain the set of linkage performance characteristics. S502: Based on the aforementioned set of linkage performance characteristics, classify and group the pathological linkage signs according to their quantity, trend inflection point concentration, and directional consistency. Call the risk response pattern category defined in the structural classification standard, compare the classification results with the structural characteristics of the established patterns, determine the risk structure label, and obtain the risk category attribution list. S503: Based on the labels listed in the risk category attribution list, label each linkage relationship in the sepsis early warning result with the corresponding risk response category to obtain the risk classification result.

Images