Multivariable trend anomaly detection method for heart failure home patient

Abstract

The invention relates to a heart failure home patient-oriented multivariable trend anomaly detection method, which comprises the following steps of: constructing an initial contour of a multivariable health trend for a patient, and introducing a trend inertia vector: updating trend inertia to form an individual trend trajectory relationship; detecting an inertia breaking point in the trend trajectory diagram; analyzing whether the variables within 12-36 hours before and after the focusing breaking point analysis generate collaborative disturbance or not; performing reprocessing through a disturbance amplification operator to form a potential early warning factor; constructing a multivariable intervention graph by using the disturbance cooperation matrix; monitoring the offset direction and strength of the causal propagation chain on each path; if the plurality of paths are subjected to direction deviation in the same period at the same time, judging that the deviation is pathological trend deviation; converting the causal offset path into a single trend risk factor, spreading a plurality of weak signals in a variable graph, evaluating systematic influence, forming a fuzzy risk scoring curved surface relationship, and outputting an individual trend risk level; the sensitivity and continuity of trend identification are improved, individual differences are adapted, and the false alarm rate is reduced.

Description

Technical Field

[0001] The present invention relates to a multivariate trend anomaly detection method, and more specifically to a multivariate trend anomaly detection method for patients with heart failure at home. Background Art

[0002] Current methods for early detection of heart failure, such as patent CN111063453A, demonstrate some innovation in incorporating big data and artificial intelligence into early detection of heart failure. However, these methods still suffer from significant deficiencies and technical limitations when applied to detecting multivariate trend anomalies in heart failure patients at home. These deficiencies are particularly pronounced in the areas of trend modeling, variable collaborative perception, adaptation to individual differences, causal interpretation, and risk warning accuracy. These shortcomings are particularly prominent in the long-term home management of chronic heart failure. First, from a data modeling perspective, existing methods primarily rely on static aggregation and similarity assessment strategies of historical physiological data. Their core logic is to use similar patterns of parameters in time series for anomaly detection. However, they fail to incorporate dynamic trend inertia modeling and the joint characterization of trend direction and persistence. Consequently, they are unable to effectively capture the structural evolutionary characteristics of continuous, slowly varying variables over time. In particular, these methods overlook trend structural features such as persistence, reversal, and mutation points. This is extremely detrimental to identifying the most clinically valuable trend signals in heart failure, such as slow deterioration or pre-collapse fluctuations. Secondly, although the method integrates physiological parameters such as respiratory rate, heart rate, blood pressure, and weight, it lacks in-depth exploration of the trend synergy among multiple variables. It only uses similarity as a measurement indicator and fails to establish a causal network or transmission path model between variables. As a result, the system cannot determine whether a change in a certain indicator is a dominant factor, a response result, or a non-causal interaction, thereby limiting the early warning explanatory power and intervention traceability capabilities. In the scenario where multiple variables fluctuate simultaneously, it may cause false alarms or ignore the real key variables.

[0003] In addition, the invention still uses a detection method based on current values ​​or local change rates in the impact assessment of variable fluctuations, ignoring the continuous expression of trend inertia. This makes it more sensitive to short-term noise, especially when patients experience non-pathological fluctuations such as body movement interference, occasional drinking, and decreased sleep quality. It is very easy for the system to misjudge it as a clinical risk, increasing the psychological burden on patients and the burden of medical response. Furthermore, although the solution proposes the use of artificial intelligence for intelligent analysis, it does not clearly describe whether to integrate the patient's individual baseline drift modeling mechanism. Therefore, its model is difficult to adapt to the slow-changing characteristics of physiological parameters during long-term monitoring of heart failure patients. For example, heart rate or weight may show a new normal trend of month-to-month changes in some patients, and traditional models may still be judged as abnormal without dynamic baseline correction, further exacerbating false alarms. In addition, although this method is mentioned as being used for early identification of decompensation events, it does not achieve early perception of pre-decompensation trends in terms of mechanism, that is, it does not introduce structural trend change detection methods (such as trend breaks, trend synergy enhancement, and trend resonance segment identification). Its logic still relies heavily on a static framework that triggers an early warning only after the indicators have changed. Therefore, it cannot meet the clinical need to identify critical trends 48 hours in advance, and misses the golden window for intervention.

[0004] From the perspective of the deployment environment, this solution is more suitable for cloud-based analysis of data from hospitals or community medical sites. It is not optimized for device performance limitations, real-time feedback, edge computing, and active patient participation in the home environment. Therefore, it is difficult to implement in terms of real-time performance, feedback interactivity, and adaptability at home. In addition, this method does not establish a closed-loop mechanism for data feedback, and users' subjective feedback (such as self-reported symptoms such as fatigue and edema) cannot participate in system learning, resulting in the model's lack of self-correction capabilities in long-term operation, making it difficult to truly achieve individualized dynamic evolution. Finally, from the perspective of multivariable system modeling, existing technologies do not distinguish between structural trend deviation behaviors between variables, nor do they provide means for assessing the intensity of trend path deviations and identifying abnormal propagation chains. When multiple parameters of a patient change simultaneously, it is impossible to distinguish whether it is a systemic disorder or an occasional disturbance, which limits its intelligent response capabilities during periods of complex and fluctuating conditions. Summary of the Invention

[0005] The purpose of the present invention is to provide a multivariate trend anomaly detection method for home patients with heart failure, thereby solving some of the drawbacks and shortcomings pointed out in the background technology.

[0006] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: a multivariate trend anomaly detection method for home-based heart failure patients, comprising: constructing an initial profile of the multivariate health trend for each patient, introducing a trend inertia vector; defining and recording the trend direction and duration of each parameter, including increase / decrease / stable; updating the trend inertia to form an individual trend trajectory relationship; and allowing the patient's health status to be identified in the form of a progressive downward spiral; Detect inertial breakpoints in trend trajectory graphs; focus on analyzing whether variables within 12-36 hours before and after the breakpoints produce coordinated disturbances; reprocess the coordinated disturbance signals through a disturbance amplification operator to form potential early warning factors; construct a multivariate intervention map using a disturbance coordination matrix, with each edge representing a single medically plausible variable impact path; monitor the offset direction and strength of the causal transmission chain on each path; if multiple paths experience simultaneous directional deviations within the same period, it is determined to be a pathological trend deviation; Each causal deviation path is converted into a single trend risk factor and assigned a fuzzy weight. Multiple weak signals are propagated in the variable graph to evaluate the systemic impact and form a fuzzy risk score surface relationship to output the individual trend risk level.

[0007] Furthermore, the trend inertia update method for forming individual trend trajectory relationships includes: Collect time series data on multiple physiological variables of patients in a home environment, including heart rate, blood pressure, weight, respiratory rate, and blood oxygen saturation; establish a trend inertia vector for each physiological variable. The trend inertia vector is used to describe the trend direction and trend persistence of the variable over a continuous time period, where the trend direction is rising, falling, or stable, and the trend persistence is the length of time that the trend direction is maintained; Connect trend inertia vectors in chronological order to construct a trend evolution sequence for the variable and record trend direction changes or trend break events; construct individual trend trajectory maps based on trend inertia vector sequences of multiple variables to identify synergy or anomalies in trend direction and trend persistence among multiple variables; According to the trend trajectory map, the individual's health status stage is identified, including the trend deterioration period, trend stability period, and trend reversal period, and serves as the input basis for subsequent trend anomaly detection and early warning analysis.

[0008] Furthermore, the trend inertia vector is generated through daily dynamic updates. When the variable trend direction changes or the trend persistence exceeds a preset stability threshold, the trend segment is triggered to terminate and a new trend inertia vector is rebuilt. The trend direction is determined based on the continuous daily change direction of the variable value, and the trend persistence is obtained by calculating the continuous time during which the trend is not interrupted.

[0009] Furthermore, the trend inertia vector information of each physiological variable within a continuous time period is extracted, and the trend direction and duration, including rising, falling, and stable, are encoded as structural units. The presence of trend inertia segments with temporal overlap and consistent direction is identified between multiple variables, and an individual trend trajectory map with trend synergy is constructed. In order to mathematically express the structural relationship and offset degree in the trend trajectory map, the following trend structure function is designed: in: : Trend structure function, indicating that at time The degree of structural coordination of the trend map; : The total number of physiological variables monitored including weight and heart rate; :variable The trend direction function is , indicating decline, stability, and rise respectively; :variable The length of time that the current trend has lasted; :variable The trend strength adjustment factor is calculated by combining the short-term fluctuation amplitude of the variable with the degree of deviation from the long-term average trend; :variable The trend time overlap factor with other variables in the current time window, with a value range of , which is used to measure the degree of trend synchronization between variables; : Trend modeling start time; : current time point; When multiple variables have the same trend direction (such as rising at the same time) in the same time period, and the trend lasts for a long time (large inertia), the trend intensity is high (obvious fluctuations), and the time overlap is strong, The value will rise rapidly; the trend trajectory map can be analyzed Growth rates, extreme points, and inflection points can help identify potential in the system: Trend turning points where trend inertia reverses; Trend breaking points where inertia is interrupted or direction is disturbed; Trend resonance segment where multiple variable trends are synergistically enhanced; The above structural changes can serve as basic data support for trend risk scoring, resonance anomaly analysis, and intervention time window recommendation.

[0010] Furthermore, the method for determining a pathological trend deviation includes: S1. Construct a multivariate causal transmission map. This map, based on the medical knowledge map and historical monitoring data, defines a directed causal chain structure between variables to reflect the influence path between variables. The trend direction and transmission strength of each causal chain are monitored in real time. When a chain is detected to have a trend reversal, trend acceleration, shortened or amplified within a given time window, it is recorded as a trend shift event. S2. If, within the same monitoring period, multiple related causal chains experience a trend shift simultaneously, and the shift direction is consistent, it is determined to be a pathological trend shift. S3. Pathological trend deviation is used as one of the health risk triggering conditions to generate a trend risk level assessment and output health warning information of the corresponding level.

[0011] Furthermore, the trend direction refers to the changing trend of the physiological variable in a continuous time period, and the trend conduction intensity refers to the response intensity or time delay change generated by the trend change of one variable being transmitted to the next variable on its influence path.

[0012] Furthermore, the trend deviation event includes any one or more of the following: The trend direction between variables in the causal chain changes from stability to synchronous strengthening; The same path changes from weak correlation to strong correlation in a short period of time; The excursion pattern that has not appeared in historical data occurs for the first time; The conduction time of the influence path between variables is significantly shortened or mutated.

[0013] Furthermore, the determination of the pathological trend deviation satisfies at least two of the following conditions: Two or more causal paths shift in direction within the same period; The offset path does not appear in the historical trend graph; The intensity of trend transmission after the deviation exceeds the established risk threshold; The shift occurred along the typical high-risk variable pathway for heart failure, including weight → respiratory rate → heart rate.

[0014] The present invention has the following beneficial effects: By recording the trend direction (increase, decrease, stability) and duration of variables using trend inertia vectors, it shifts from point-value analysis to trend segment identification. This allows for capturing slowly accumulating risk changes in heart failure patients on a daily basis, and is particularly suitable for identifying early pathological trends that are not dramatic but long-term. By constructing individual trend trajectory maps and analyzing the consistency and synergy of trends across multiple variables, it not only determines whether a particular indicator is abnormal but also identifies whether there are structural offsets between multiple parameters, thereby providing early warning of system-level health imbalances and avoiding false positives and missed alerts.

[0015] Based on the clinical medical knowledge graph, a directed causal chain is established between variables, and trend deviation events along the causal path, such as trend direction reversal, response acceleration, and shortened conduction lag, are monitored in real time. This allows the system to construct risk models based on actual pathological mechanisms, rather than relying on static statistical correlations. By comprehensively evaluating trend synergy, trend strength, and conduction velocity through structural functions, the system can identify trend turning points, trend breakpoints, and trend resonance segments, thereby predicting risk outbreak points in advance and providing patients or doctors with clear time windows for intervention recommendations. This adapts to individual differences and reduces false alarm rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of multivariate trend inertia analysis and risk warning of the present invention.

[0017] Figure 2 This is a functional relationship diagram of multivariate trend inertia and intervention path of the present invention.

[0018] Figure 3 This is a relationship diagram between the fuzzy diffusion of trend risk factors and the scoring surface of the present invention.

[0019] Figure 4 This is a flow chart of continuous home monitoring and trend warning for heart failure patients according to Example 1 of the present invention.

[0020] Figure 5 This is a closed-loop structure diagram of multivariate trend deviation determination and risk warning in Example 2 of the present invention. DETAILED DESCRIPTION

[0021] The following is a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings.

[0022] Combined with attachment Figure 1The present invention is a multivariate trend anomaly detection method for patients with heart failure at home. It establishes an individualized multivariate health trend initial profile for each patient, and obtains it by dynamically modeling a number of core physiological parameters continuously collected in the patient's daily home environment. The parameters include but are not limited to weight, heart rate, systolic blood pressure, diastolic blood pressure, respiratory rate and blood oxygen saturation, etc. Each parameter is collected and normalized in a time series manner. The modeling method of introducing trend inertia vector is used to replace the traditional anomaly identification mechanism based on single-point judgment. Trend inertia vector is a structural expression method for dynamically describing the trend state of a variable. Its core lies in jointly modeling the trend direction (defined as rising, falling or stable) of each parameter in a continuous time period and the duration of the trend (that is, the natural time length that the trend direction continues without interruption) and using this as the basic unit of the trend evolution state. Based on the daily change direction of each variable, the system calculates its current trend inertia state in real time. When a trend direction remains unchanged, the duration of trend inertia accumulates over time. When the trend direction reverses or fluctuates dramatically, the previous trend inertia segment terminates and a new inertia vector segment is reconstructed, thus forming a time series of continuous trend inertia vector segments. By fusing and analyzing the sequence information of these multivariate trend inertia vectors, an individual trend trajectory map with temporal structure and trend continuity is constructed. This map is used to characterize the evolution of a patient's health status along the temporal dimension. Instead of relying on the instantaneous fluctuations of a single variable to determine health risk, it instead divides trend stages based on the overall deviation of the trend structure. Unlike traditional outlier detection or threshold judgment methods, this method allows the system to capture the phenomenon of increasing trend consistency across multiple variables in a patient. It can identify early-stage states characterized by slow deterioration and synchronous deviation of the overall trend, even though individual indicators have not yet reached abnormal thresholds. This is particularly applicable to chronic diseases such as heart failure, where the progression of the disease often manifests as a long-term latent and gradual accumulation. During continuous observation by the system, if the trend inertia of multiple parameters continues to decline (i.e., the health of the indicators deteriorates) and the duration is expanding, and at the same time, the trend turning points become more frequent or the trend inertia interruptions become more severe, the system can determine the trend trajectory as a progressive downward spiral state, that is, the patient's health status has entered a trend deterioration state from a stable state but has not yet triggered the traditional acute alarm mechanism, thereby achieving the purpose of early identification and response to pathological risks.

[0023] Combined with attachment Figure 2Based on the individual trend trajectory map of patients constructed in the early stage, the system first performs a structured analysis of the trend inertia sequence in the trend trajectory map to identify the inertia breakpoints. The inertia breakpoint refers to the node where the trend inertia vector of a certain physiological variable undergoes a sudden change, interruption or reversal in trend direction or continuity, usually accompanied by a break in trend stability or a significant increase in trend volatility. After detecting the inertia breakpoint, the system will use the breakpoint as the time anchor point and extend the time window of 12-36 hours forward and backward respectively, extract the trend inertia state of other physiological variables in this time period, and calculate whether these variables have trend collaborative disturbances within the window, that is, multiple variables have collaborative deviations such as trend direction convergence, trend acceleration or inertia interruption within the time overlap range, forming a preliminary collaborative disturbance candidate segment. The system inputs the aforementioned collaborative disturbance signal into a disturbance amplification operator, a mapping function designed to enhance the semantic strength of weak trend signals, highlighting trend anomalies such as trend consistency, suddenness, and high synergy. Through nonlinear weight adjustment and time alignment, the original trend disturbance signal is reprocessed and a normalized potential early warning factor is generated. Furthermore, based on the medical knowledge logic and historical synergy data between variables, the system constructs a multivariate intervention map. The intervention map is a directed graph structure, in which each node represents a physiological variable and each edge represents a single medically plausible variable influence path. For example, changes in weight can affect respiratory rate, which in turn can affect heart rate fluctuations. The system treats these paths as potential causal transmission chains. Within the intervention map, the system continuously monitors the direction and intensity of trend transmission between variables along each path. Changes in trend transmission direction refer to whether the trend influence between variables has reversed, synergized, or disconnected. Changes in trend transmission intensity refer to whether the response of the influence in downstream variables has significantly increased or occurred earlier. When the system detects that two or more paths in the intervention map simultaneously shift in trend direction within the same time period, and the impact intensity increases significantly or shows an increasing trend of concentration, the system marks the structural change of the trend as a pathological trend shift, that is, the patient's current state has entered a systemic trend disorder state from a stable trend state, which usually indicates a critical stage of potential rapid deterioration of health or imbalance of functional regulation, thereby triggering the subsequent intelligent risk assessment module to generate early warning indicators, providing patients and caregivers with advance intervention suggestions or professional guidance.

[0024] Combined with attachment Figure 3Based on the continuous monitoring results of individual trend trajectory maps and intervention path maps, after identifying the existence of pathological trend deviations, the system further performs structural transformation on all causal paths with significant trend deviations, and extracts each deviation path as an independent trend risk factor. The trend risk factor comprehensively considers the deviation degree of the starting variable of the path, the frequency of the trend inertia rupture in the path, the suddenness of the response of the endpoint variable, and the importance weight of the path in the entire map, and preliminarily quantifies its risk degree through the structural scoring model. Then the system assigns a fuzzy weight label to each trend risk factor. The fuzzy weight is not a fixed risk level, but a dynamic fuzzy membership value calculated based on the current deviation intensity, historical variation range and individual characteristics of the patient, which is used to represent the true intervention urgency or trend impact possibility of the risk factor. The system deploys all trend risk factors, in their weighted form, within a previously constructed variable impact graph and activates a fuzzy diffusion mechanism, executing a weak signal propagation process within the graph structure. This means that multiple risk factors are no longer calculated individually, but rather interact within a multivariate network. Fuzzy state diffusion functions between graph nodes simulate the path transmission and superposition of potential impacts, assessing the comprehensive impact of trend perturbations on the system. A fuzzy trend risk score surface is generated for the current time slice, reflecting the distribution density and transmission strength of trend risk across variables. The individual trend risk level at the current moment is calculated using the local peaks, edge mutations, and global average gradient of the risk surface. Finally, based on the assessment results of the fuzzy risk score surface, the system outputs the patient's current health status as a multi-level trend risk label, such as normal, mild fluctuation, suspicious deviation, high-risk deviation, and severe warning level. This label is used to guide patients and their caregivers in targeted health interventions or report to the medical system for manual review, thus achieving a closed-loop risk output mechanism for continuous trend risk modeling, weak signal integration, system propagation analysis, and fuzzy intelligent early warning.

[0025] Example 1: Combined with attachment Figure 4In this embodiment, a 68-year-old male heart failure patient, Mr. Zhang, continuously monitors six core physiological parameters during his home rehabilitation phase through wearable devices and smart home terminal systems, namely heart rate, systolic blood pressure, diastolic blood pressure, weight, respiratory rate and blood oxygen saturation. The system records them once a day and automatically uploads and synchronizes them every week. The following are some monitoring data of Mr. Zhang for the past 10 days: Day 1: heart rate 74 beats / min, weight 69.5kg, systolic blood pressure 122mmHg, respiratory rate 17 beats / min, blood oxygen 97%; Day 2: heart rate 76, weight 69.8kg, systolic blood pressure 124, respiratory rate 18, blood oxygen 96%; Day 3: heart rate 78, weight 70.1kg, systolic blood pressure 126, respiratory rate 18, blood oxygen 96%; Day 4: heart rate 80, weight 70.5kg, systolic blood pressure 129, respiratory rate 19, blood oxygen 95%; Day 5: heart rate 82, weight 70.9kg, systolic blood pressure 132 , respiratory rate 19, blood oxygen 95%; Day 6: heart rate 85, weight 71.3 kg, systolic blood pressure 134, respiratory rate 20, blood oxygen 94%; Day 7: heart rate 86, weight 71.5 kg, systolic blood pressure 135, respiratory rate 20, blood oxygen 94%; Day 8: heart rate 87, weight 71.8 kg, systolic blood pressure 136, respiratory rate 21, blood oxygen 94%; Day 9: heart rate 88, weight 72.0 kg, systolic blood pressure 137, respiratory rate 22, blood oxygen 93%; Day 10: heart rate 89, weight 72.3 kg, systolic blood pressure 139, respiratory rate 22, blood oxygen 93%. The system performs trend analysis on each physiological data item over these 10 days and extracts trend inertia vectors. For example, heart rate shows a continuous upward trend from day 1 to day 10, with an upward trend and a duration of 10 days. Weight also shows an upward trend with a duration of 10 days. Systolic blood pressure rises from 122 to 139, also with an upward trend, and a duration of 10 days. Respiratory rate rises from 17 to 22, with an upward trend and a duration of 10 days. However, blood oxygen saturation slowly decreases from 97% to 93%, and the system determines that its trend direction is downward, with a duration of 10 days. As a result, the system generates the following trend inertia vector group for Mr. Zhang: {heart rate: rising - 10 days, weight: rising - 10 days, systolic blood pressure: rising - 10 days, respiratory rate: rising - 10 days, blood oxygen: falling - 10 days}. Each inertia vector also records the trend growth amplitude, the average daily rate of change, and whether there are fluctuation interruptions. Subsequently, the system arranges the above trend inertia vectors in chronological order to form a trend evolution sequence, and searches for trend breakpoints or trend direction change points in each variable. After calculation, there are no breakpoints in the current stage, and the trend continuity is strong. The system marks it as a highly stable and continuous trend segment.Furthermore, the system conducts a cross-variable synergistic analysis of all trend inertia vectors to construct a trend trajectory map. In this map, all variables exhibit consistent temporal overlap and trend synergy. Specifically, multiple variables exhibit the same trend direction (four rising, one falling) and nearly identical trend durations (all 10 days). The system calculates a synergy factor of 0.93, indicating a highly consistent synergistic trend. Based on the structural deduction rules for the trend trajectory map, the system identifies this state as a period of trend deterioration. This determination is based on the fact that multiple physiological parameters exhibit trend inertia toward pathological progression, such as a sustained increase in heart rate, respiratory rate, blood pressure, weight gain (indicating potential fluid retention), and a sustained decrease in blood oxygen, consistent with the typical process of progressive decompensation of heart failure. According to the system's pre-set rules, if the trend inertia synergy is greater than 0.85, lasts for more than 7 days, and involves three or more high-weighted indicators, the system enters the trend anomaly warning analysis process. The system pushed this trend trajectory map to the trend risk scoring engine, linking it with subsequent disturbance identification, causal shift analysis, and fuzzy risk scoring modules. It also automatically generated a trend briefing within the patient's home app, indicating the early stages of a worsening trend and recommending enhanced monitoring and assessment of fluid intake and medication compliance. Following the system prompt, Mr. Zhang's caregiver contacted his community doctor for guidance and intervention, ultimately avoiding a short-term hospitalization for an acute exacerbation caused by the accumulated trend.

[0026] After his condition entered a period of worsening, the system continued to automatically collect his daily physiological data and dynamically update the trend inertia vector. From the 11th to the 15th day, Mr. Zhang's weight dropped from 72.3 kg to 71.2 kg, his heart rate dropped from 89 to 86 beats per minute, his systolic blood pressure dropped from 139 to 131 mmHg, his respiratory rate dropped from 22 to 19 breaths per minute, and his blood oxygen saturation gradually recovered to 95%. The system activates the trend update mechanism based on daily data changes, and re-evaluates and calculates the trend direction and continuity of each variable. The following is the specific data and trend analysis process: On the 11th day, weight 71.9kg, heart rate 88, systolic blood pressure 137, respiration 21, blood oxygen 94%; on the 12th day, weight 71.7kg, heart rate 87, systolic blood pressure 135, respiration 20, blood oxygen 94%; on the 13th day, weight 71.5kg, heart rate 86, systolic blood pressure 133, respiration 19, blood oxygen 95%; on the 14th day, weight 71.3kg, heart rate 86, systolic blood pressure 132, respiration 19, blood oxygen 95%; on the 15th day, weight 71.2kg, heart rate 86, systolic blood pressure 131, respiration 19, blood oxygen 95%. After analyzing the weight variable, the system discovered that weight had been declining for three consecutive days, reversing the direction of the previous 10-day upward trend. Since this trend had been uninterrupted and had lasted for three days, satisfying the trend direction change and exceeding the stability switching threshold (two days in this embodiment), the original "Upward - 10 Days" trend segment was terminated by the system, and a new trend inertia vector, "Downward - 3 Days," was generated. The heart rate variable was updated from "Upward - 10 Days" to "Downward - 3 Days." On days 13, 14, and 15, the heart rate remained at 86. Since there was no significant change within three days, the system considered a heart rate fluctuation of less than 1 bpm to be stable. Therefore, the system configured the segment to terminate after "Downward - 2 Days" and generate "Stable - 1 Day." The systolic blood pressure dropped from 139 to 131, forming a new trend inertia segment, "Downward - 5 Days." The respiratory rate dropped from 22 to 19, continuing to decline for more than three days, resulting in a "Downward - 4 Days" segment. The blood oxygen level rose from a minimum of 93% to 95%, forming an inertia segment, "Upward - 3 Days." The system replaced the original 10-day trend segment with the above-mentioned trend inertia segment, forming a new set of trend segment nodes in Mr. Zhang's trend trajectory map. At the same time, the system compared the directional changes of the new and old trend segments, marked the variables with reversed direction (such as weight and blood pressure) as trend reversal points, and marked the variables that re-entered a stable state after the trend terminated (such as heart rate) as trend buffer segments, which are used for subsequent trend stage judgment and abnormal evolution chain identification. Collaborative analysis found that the trend direction of the three variables of weight, respiration, and blood pressure consistently changed from rising to falling, and the trend time was synchronized for more than 3 days. The system determined that it was a collaborative trend reversal, and this reversal trend reflected that the intervention measures were taking effect.The risk engine lowers the individual trend risk score level from a high-risk deviation to a suspicious deviation stabilization, and uploads the trend change to the trend learning module to provide label data for model optimization.

[0027] From the 16th to the 22nd day, the system continued to dynamically track its multivariate trend changes and introduced the trend structure function for the first time. Conduct structural trend map evaluation and abnormal structure identification.

[0028] The key variable data from the 16th to the 22nd are as follows: body weight (kg): 71.0, 70.8, 70.7, 70.7, 70.9, 71.2, 71.5; heart rate (bpm): 86, 85, 84, 85, 86, 87, 88; systolic blood pressure (mmHg): 130, 130, 132, 134, 136, 137, 139; respiratory rate (b / min): 18, 18, 19, 20, 21, 21, 22; blood oxygen (%): 95, 95, 94, 94, 94, 93, 93. Based on the results of the daily data trend direction changes, the system identified the following trend inertia vectors: weight decreased from the 16th to the 18th day (D=–1, L=3), stabilized on the 19th day (D=0, L=1), and increased from the 20th to the 22nd day (D=+1, L=3); heart rate continued to increase (D=+1, L=4); systolic blood pressure continued to increase (D=+1, L=7); respiratory rate continued to increase (D=+1, L=5); blood oxygen continued to decrease (D=–1, L=4).

[0029] The system calculates the trend strength adjustment factor for each variable , which is defined as: Setting coefficient 、 , with a value range of [0.1, 1.0]. Taking weight as an example, the daily changes from day 20 to day 22 are +0.3kg, +0.3kg, +0.3kg, with a short-term volatility of 0.3kg and a long-term deviation from the mean (baseline is 70.5kg) of +1kg. The relative deviation rate is 1.0 / 70.5≈0.014. Other variables such as heart rate, respiratory rate, blood pressure, etc. are calculated in the same way. The values ​​were 0.22, 0.25 and 0.30 respectively, and the blood oxygen level decreased. .

[0030] Next, the system calculates the trend time overlap factor between variables Calculation is performed. If the trend direction of the two variables is consistent in the current window and the overlap ratio of the trend duration days is greater than 80%, then The maximum value is 1.0. If the trends are different or only partially overlapped, the linear weight reduction value range is [0.3, 0.9]. In this stage, the heart rate, systolic blood pressure and respiratory rate all show an upward trend, and the time overlap is more than 90%, so , weight due to the trend reversal, its rising segment , blood oxygen decreases in the opposite direction, set .

[0031] Substituting the above values ​​into the trend structure function: Take the 22nd day as an example ( =Day 22, = 16th day), the current trend inertia segment is calculated as follows: weight: , , , →Component value Heart rate: , , , →Component value Systolic blood pressure: , , , →Component value Respiratory rate: , , , →Component value Blood oxygen: , , , →Component value Total integrand value ≈ For this value =16 to =22, and the discrete summation approximate integral is assumed during the day. , is the current trend structure function value.

[0032] The reference range of the system-set risk threshold is as follows: The 22nd day marked Mr. Zhang's structure function value as 1.18, which was determined to be in the trend resonance enhancement zone and marked by the system as a synergistic worsening resonance segment. Due to its consistent trend direction, concentrated inertia, and significant temporal overlap, the system predicted that more dramatic parameter fluctuations would occur within the next 2-3 days based on the growth trend of the structure function, triggering an early warning mechanism and recommending a review of heart failure symptoms, self-weight records, and medication compliance within 48 hours. The family also received a notification indicating that the patient had entered the critical excursion period.

[0033] Example 2: Combined with attachment Figure 5 Based on Example 1, during the 23rd to 28th day of Mr. Zhang's home health monitoring, the system discovered that several of his physiological parameters showed subtle but synchronous trend changes. The data from the 23rd to the 28th day were as follows: weight (kg) 71.7 → 72.2, heart rate (bpm) 88 → 92, systolic blood pressure (mmHg) 139 → 144, respiratory rate (bpm) 22 → 24, blood oxygen saturation (%) 93 → 91. The system performs the following three steps in sequence based on the pathological trend deviation method: S1 constructs a causal propagation map and monitors trend deviation events: The system first constructs Mr. Zhang's personalized multivariate causal propagation map based on previous historical data learning and the heart failure clinical pathway knowledge base. The nodes in the map represent variables (such as weight, respiratory rate, heart rate, etc.), and the edges represent causal relationships. For example, weight increases → respiratory rate increases (indicating that fluid retention causes pulmonary congestion), respiratory rate increases → heart rate increases (indicating increased oxygen consumption induces sympathetic excitement), systolic blood pressure increases → heart rate increases (representing the compensatory reaction of the heart's pumping force after increased vascular resistance), which equivalently constitutes a directed chain: weight ↑ → respiratory ↑ → heart rate ↑, and systolic blood pressure ↑ → heart rate ↑. The system monitors the direction and intensity of trend propagation along these chains daily. For example, from the 24th to the 28th day, the three variables showed a synchronous increase in the trend direction, and the trend lasted for five consecutive days. The conduction delay was shortened from an average of two days to one day (i.e., faster breathing and heart rate occurred the day after weight increased). Furthermore, the conduction intensity between systolic blood pressure and heart rate was significantly enhanced, with the daily increase in systolic blood pressure increasing from 0.5 mmHg to 1.2 mmHg, and the heart rate response showing a significant jump either synchronously or the next day (for example, on the 27th day, systolic blood pressure rose from 142 to 144, and heart rate rose from 90 to 92). The system determined that the response delay of this chain was shortened and the impact was deepened, meeting the criteria for trend direction shift and conduction enhancement. Therefore, the trend shift event was recorded separately on each path.

[0034] S2: Synchronous shifts in multiple chains were identified as pathological trend shifts. Next, the system performed a structural cascade analysis of these trend shift events and discovered that at least two core chains (weight → respiration → heart rate, systolic blood pressure → heart rate) experienced a sustained upward trend within the same cycle (days 24 to 28). This trend lasted for ≥3 days, and the trend acceleration (the second-order difference of the magnitude of change) was positive, indicating not only an increase but also an accelerated increase, and the shift direction was consistent (all deteriorating). Based on the system's set criteria, if two or more high-risk chains experience simultaneous trend shifts in the same direction for more than three days, it constitutes a pathological trend shift. Therefore, Mr. Zhang's trend change during this cycle was officially identified by the system as a pathological trend shift.

[0035] S3 outputs the trend risk level and issues a health warning: the system inputs the above judgment results into the trend risk scoring engine and calls the above trend structure function , and according to its growth rate (this cycle The increase is +0.95), matching the risk surface model, and outputting a trend risk level of high-risk deviation. The risk label pathological co-deviation is bound to Mr. Zhang's user profile and triggers a home app early warning notification, indicating that the system has detected a synchronous increase in the trend of key indicators of heart failure, which has constituted a co-pathological trend deviation. It is recommended to complete a review of medication records and self-check salt and water intake within 48 hours, and confirm with the community doctor whether the diuretic regimen needs to be adjusted. This information is also pushed through the family side, indicating that the current status is not an occasional fluctuation but a trend evolution. After Mr. Zhang's family contacted the doctor, the diuretic dosage was adjusted, and his respiratory rate and heart rate gradually decreased after his weight stabilized. The system then recorded the trend reversal, and the risk level was gradually lowered.

[0036] During the 29th to 35th day of Mr. Zhang's home health monitoring, the system continued to implement a joint monitoring method for trend direction and trend transmission strength, further validating the structural link characteristics of pathological trend excursion. Previously, Mr. Zhang had been identified by the system as entering a pathological trend excursion due to weight gain and a coordinated deterioration in respiratory rate and heart rate. With the intervention of a physician, he adjusted his diuretic dosage and hydration management strategy. To assess the effectiveness of the intervention and prevent trend rebound, the system continuously calculated and tracked the multivariate trend direction and trend transmission strength daily. From the 29th to the 35th day, Mr. Zhang's key physiological variables were recorded as follows: weight decreased daily from 72.2 kg to 71.1 kg, respiratory rate decreased from 24 breaths / minute to 20 breaths / minute, heart rate decreased from 92 to 86 beats / minute, systolic blood pressure decreased from 144 to 135 mmHg, and blood oxygen level recovered from 91% to 94%. The system first analyzes the trend direction of each variable. Weight showed a continuous downward trend, with a duration of seven days. Respiratory rate and heart rate also showed a continuous downward trend, with consistent trend directions and inertia durations of six and five days, respectively. Blood pressure decreased for five days, while blood oxygen levels increased for four days. These trends do not conflict with each other, but indicate a stabilization of overall health. After identifying these trend directions, the system then calculates the strength of the trend transmission between these variables, including the effect of weight change on respiratory rate, the effect of respiratory rate on heart rate, and the response of the systolic blood pressure to heart rate pathway. For example, during the period from weight to respiratory rate when Mr. Zhang's weight dropped from 72.0 kg to 71.5 kg, his respiratory rate decreased from 23 to 21 within 48 hours, demonstrating a short delay and rapid response. The system therefore labeled this path's trend transmission strength as strong. When his weight dropped to 71.1 kg, his respiratory rate also decreased to 20, indicating that the trend relationship not only persisted but also showed a synchronous strengthening. Similarly, the trend relationship between respiratory rate and heart rate showed consistent direction and a timely response. Within one day of each decrease in respiratory rate, heart rate decreased accordingly, indicating that heart rate is highly sensitive to changes in respiratory rate, and the strength of the trend transmission was assessed as moderate to strong. Although the downward trend in systolic blood pressure was consistent with the direction of the decrease in heart rate, the fluctuation in heart rate was relatively smaller than the decrease in systolic blood pressure throughout the observation period, and there was a certain delay on some days, indicating that the transmission strength was weak to moderate, which is controlled by the combined influence of other drugs or individual regulatory mechanisms.

[0037] During the trend conduction intensity modeling process, the system not only records whether the trend directions between variables are consistent, but also determines whether these changes are directly related or just synchronous phenomena through the duration, response amplitude and offset rate of the variable response. Finally, based on the consistency of trend direction and the distribution of trend conduction intensity, the system updates the edge weights of each path in the trend map, and marks the trend paragraph structure as a reverse collaborative trend repair period, indicating that multiple key variables of the current patient are recovering in a healthy direction, but it is necessary to continue to observe whether there is a new round of reversal in its trend. Based on the linkage between the stability of the trend direction and the dynamic changes in the conduction intensity in the conduction path, the system outputs a trend improvement stage score of medium to high, and generates a health feedback briefing to the patient and his caregiver. After a remote check, the doctor further confirmed that Mr. Zhang's fluid retention control was stable, decided to maintain the current drug dosage and recommended continued daily monitoring.

[0038] After Mr. Zhang entered the monitoring period from day 36 to day 42, the system again detected complex changes in multiple physiological indicators. Based on the trend shift event definition framework, the system began to conduct in-depth analysis of trend behaviors within the causal chain structure to determine whether new pathological evolution signals were emerging. Mr. Zhang's monitoring data during this period were as follows: weight rebounded from 71.2 kg to 72.1 kg, respiratory rate increased from 20 breaths / minute to 23 breaths / minute, heart rate increased from 86 to 91 beats / minute, systolic blood pressure increased from 135 to 142 mmHg, and blood oxygen level decreased from 94% to 91%. Based on the daily updates of the trend inertia vector, the system identified that Mr. Zhang's weight began to increase from day 37, for six consecutive days, with daily increases ranging from 0.15 kg to 0.2 kg. Simultaneously, respiratory rate began to accelerate from day 38 to day 42, heart rate saw a significant jump on day 39, systolic blood pressure also reversed from a slow decline to a rapid recovery, and blood oxygen level showed a slight downward trend, initially suggesting the formation of a coordinated trend shift among multiple variables. The system first analyzes the causal chain to see if there's a shift in trend direction from stability to simultaneous strengthening. From days 33 to 36, Mr. Zhang's weight, respiration rate, and heart rate remained relatively stable, with changes below the daily fluctuation threshold. However, starting on day 37, all three variables entered an upward trend almost simultaneously: weight increased by 0.6 kg, respiration rate increased by 3 breaths per minute, and heart rate jumped by 5 bpm. This shift not only shifted the trend direction from stable to rising, but also simultaneously strengthened. The system identified this as a Type I trend shift event: a shift from stable to simultaneous strengthening. The system then examined the weight → respiration → heart rate path structure and found that the correlation coefficient for this path in the previous cycle was approximately 0.42, which is considered low to medium. However, from days 37 to 42, the simultaneous changes in these three indicators increased the correlation between the paths to 0.79, which is considered a high correlation. This shift occurred within just six days. The system identified this rapid shift as a Type II trend shift event: a shift from a weak correlation to a strong correlation within the same path within a short period of time. The system then compared the current path change with Mr. Zhang's entire trend trajectory over the previous two months. It discovered that while weight gain and increased breathing had previously occurred, there had never been a coordinated trend shift pattern: a simultaneous increase in weight, breathing, heart rate, and blood pressure, along with a decrease in blood oxygen. This combination was unprecedented in Mr. Zhang's individual history, and the system recorded it as a Category III trend shift event: the first occurrence of a shift pattern not previously seen in historical data. Furthermore, the system specifically monitored the transmission time of the trend-influencing pathways between various variables.In the previous cycle, the average lag between weight changes and respiratory rate was about 2 days. However, between the 37th and 39th days, respiratory rate increased significantly only one day after weight increased, and the lag time was shortened to 1 day. Similarly, after the respiratory rate increased, the heart rate only jumped within 1 day, and the blood pressure response also quickly reversed and increased. The system determined that the above path had a significant time delay shortening through the delayed sliding window model, and classified it as the fourth type of trend shift event: the conduction time of the influencing path between variables was significantly shortened or mutated.

[0039] In summary, the system identified four types of trend deviation events during this phase, forming a trend deviation event cluster. The system triggered the trend structural risk assessment process based on the event accumulation principle. Based on the number of deviation events, synergy, first occurrence, and path strength changes, the risk scoring engine assessed Mr. Zhang's current state as a period of high trend instability, marking him as a highly sensitive segment for structural deviation. It also immediately generated a trend anomaly briefing, indicating that the patient's current physiological system is transitioning from a stable to a deteriorating state, with an accelerating rate of change and a complex structure. A comprehensive fluid assessment and medication adjustment consultation are recommended within 48 hours, and family members will simultaneously receive high-priority risk alerts.

[0040] After Mr. Zhang's home health management program entered the 43rd to 49th day phase, the system conducted a comprehensive structural analysis of his continuous trend trajectory and found that the new trend change highly met the criteria for a pathological trend shift. Previously, Mr. Zhang had experienced an increase in covariates between days 36 and 42, which was recorded as a trend shift event cluster. After medication adjustment, some indicators briefly declined. However, after day 43, the system began to identify more significant trend structural imbalances. In particular, within the typical high-risk causal pathway for heart failure (weight → respiratory rate → heart rate), the trend direction, intensity, and historical structure showed a systematic shift. Specific data showed that from day 43 to day 49, Mr. Zhang's weight increased continuously from 72.1 kg to 73.4 kg, with a daily change of +0.2 kg to +0.25 kg. His respiratory rate increased from 23 to 26 beats / minute, his heart rate increased from 91 to 95 beats / minute, his systolic blood pressure increased from 142 to 145 mmHg, and his blood oxygen level decreased continuously from 91% to 89%. The system's daily trend inertia vector updates revealed that while weight increased for seven consecutive days, respiratory rate and heart rate both began to show a synchronized upward trend starting on the 44th day, with consistent trends and an inertia segment exceeding five days, forming a trend synergy. Meanwhile, systolic blood pressure, while fluctuating slightly, continued to rise, while blood oxygen levels continued to decline throughout the entire cycle. The system extracted the weight → respiratory rate → heart rate pathway and the systolic blood pressure → heart rate pathway from the trend trajectory map and analyzed them as high-frequency conduction chains.

[0041] First, the system determines whether there are two or more causal paths that have directional deviations in the same cycle based on the first judgment condition. During this cycle, the weight → respiration → heart rate path and the systolic blood pressure → heart rate path both reversed from the previous downward / stable trend to a continuous upward trend, with the same direction and a duration of more than 3 days, meeting the trend direction deviation standard. The system marks the number of path deviation events as 2, and condition one is met. Secondly, the system reviews the structural morphology of the current path in the historical trend map. In the past 90 days of monitoring history, there has never been a complete trend structure in which weight, respiration, heart rate, and systolic blood pressure all increased simultaneously, and blood oxygen levels decreased synchronously during this stage, forming a typical pathological signal combination of increased internal pressure load + decreased gas exchange capacity. Therefore, this deviation path structure has never appeared in Mr. Zhang's historical trajectory, and it is the first abnormal path structure to occur, meeting the second condition that the deviation path does not appear in the historical trend map.

[0042] Third, the system quantifies the strength of trend transmission. For example, the weight-to-respiratory pathway previously had an average lag of 2.5 days. However, in this cycle, a one-day weight gain triggered an increase in respiratory rate the following day, shortening the lag to one day and accelerating the response. The system measured an increase in the transmission strength score from 0.38 to 0.72, exceeding the built-in risk threshold of 0.65. Similarly, the response lag for the respiratory rate-to-heart rate pathway shortened to within 24 hours, and the strength increased from 0.40 to 0.77, both exceeding the established threshold. This fulfills the third condition, indicating that the trend transmission strength after the shift exceeds the risk threshold. Finally, the shift path clearly occurs within the designated high-risk variable pathway for heart failure: weight-to-respiratory rate-to-heart rate. This pathway has high clinical interpretability and intervention value in heart failure patients, particularly given that weight gain indicates fluid retention, respiratory rate increase indicates increased pulmonary pressure, and heart rate increase indicates overactivation of cardiac compensatory mechanisms, forming a risk progression chain, thus meeting the fourth condition.

[0043] Based on the above four analysis results, the system, based on its pre-defined pathological trend excursion determination logic, determined that Mr. Zhang met at least two of the triggering criteria and all four, indicating a high-level structural trend excursion. The system immediately upgraded the trend risk level to high alert, generated a structural excursion risk map, and marked the excursion path on the trend map, triggering a doctor notification. The system identified Mr. Zhang's weight gain for seven consecutive days, which had been transmitted to his respiratory system and heart rate. The path structure deviated significantly from historical data, with a significantly increased transmission intensity, forming a typical heart failure excursion chain. The doctor was advised to assess within 24 hours whether he was experiencing fluid overload or worsening cardiac function. Simultaneously, the doctor received a graphical trend risk notification with guidance: Please monitor your fluid intake and urination over the past two days and assess for symptoms such as lower limb edema and nighttime shortness of breath. If these symptoms occur, contact your doctor immediately. Upon review, the doctor confirmed that Mr. Zhang's recent salty diet and three missed diuretic doses were due to fluid retention, leading to the worsening of his symptoms. Immediate adjustments were made to his medication and hydration regimen, successfully reversing the trend within the next five days.

[0044] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A multivariate trend anomaly detection method for patients with heart failure at home, characterized by The following steps are involved: For each patient, an initial profile of multivariate health trends is constructed, and a trend inertia vector is introduced: the trend direction and duration of each parameter, including increase / decrease / stability, is defined and recorded; Trend inertia updates form individual trend trajectory relationships; allowing patients’ health status to be identified in the form of a progressive downward spiral; Detect inertia breakpoints in trend trajectory graphs; Focus on analyzing whether the variables within 12-36 hours before and after the breakpoint produce coordinated disturbances; The collaborative disturbance signal is reprocessed through a disturbance amplification operator to form a potential early warning factor. A multivariate intervention map is constructed using a disturbance synergy matrix, with each edge representing a single medically plausible variable impact path. The deviation direction and strength of the causal transmission chain on each path are monitored. If multiple paths simultaneously deviate in direction within the same period, it is determined to be a pathological trend deviation. Each causal deviation path is converted into a single trend risk factor and assigned a fuzzy weight. Multiple weak signals are propagated in the variable graph to evaluate the systemic impact and form a fuzzy risk score surface relationship to output the individual trend risk level.

2. The multivariate trend anomaly detection method for home patients with heart failure according to claim 1 is characterized in that The method for updating the trend inertia to form an individual trend trajectory relationship includes: Collect time series data on multiple physiological variables of patients in a home environment, including heart rate, blood pressure, weight, respiratory rate, and blood oxygen saturation; establish a trend inertia vector for each physiological variable. The trend inertia vector is used to describe the trend direction and trend persistence of the variable over a continuous time period, where the trend direction is rising, falling, or stable, and the trend persistence is the length of time that the trend direction is maintained; Connect trend inertia vectors in chronological order to construct a trend evolution sequence for the variable and record trend direction changes or trend break events; construct individual trend trajectory maps based on trend inertia vector sequences of multiple variables to identify synergy or anomalies in trend direction and trend persistence among multiple variables; According to the trend trajectory map, the individual's health status stage is identified, including the trend deterioration period, trend stability period, and trend reversal period, and serves as the input basis for subsequent trend anomaly detection and early warning analysis.

3. The multivariate trend anomaly detection method for home patients with heart failure according to claim 2 is characterized in that The trend inertia vector is generated through daily dynamic updates. When the variable trend direction changes or the trend persistence exceeds a preset stability threshold, the trend segment is triggered to terminate and a new trend inertia vector is rebuilt. The trend direction is determined based on the continuous daily change direction of the variable value, and the trend persistence is obtained by calculating the continuous time during which the trend is not interrupted.

4. The multivariate trend anomaly detection method for home patients with heart failure according to claim 3 is characterized in that The trend trajectory map is used to identify multiple variables with the same direction of trend inertia vectors appearing in similar time windows, and the trend persistence is overlapping or close; The trend structure information provided by the trend inertia vector is used to assist in identifying trend turning points, trend anomaly points, and trend breakpoints, and provides a trend basic data structure for trend risk identification, trend resonance analysis, and intervention models.

5. The multivariate trend anomaly detection method for home patients with heart failure according to claim 1 is characterized in that The method for determining pathological trend deviation includes: S1. Construct a multivariate causal transmission map. This map, based on the medical knowledge map and historical monitoring data, defines a directed causal chain structure between variables to reflect the influence path between variables. The trend direction and transmission strength of each causal chain are monitored in real time. When a chain is detected to have a trend reversal, trend acceleration, shortened or amplified within a given time window, it is recorded as a trend shift event. S2. If, within the same monitoring period, multiple related causal chains experience a trend shift simultaneously, and the shift direction is consistent, it is determined to be a pathological trend shift. S3. Pathological trend deviation is used as one of the health risk triggering conditions to generate a trend risk level assessment and output health warning information of the corresponding level.

6. The multivariate trend anomaly detection method for home patients with heart failure according to claim 5 is characterized in that The trend direction refers to the changing trend of the physiological variable in a continuous time period, and the trend conduction intensity refers to the response intensity or time delay change caused by the trend change of one variable being transmitted to the next variable on its influence path.

7. The multivariate trend anomaly detection method for home patients with heart failure according to claim 6 is characterized in that The trend deviation event includes any one or more of the following: The trend direction between variables in the causal chain changes from stability to synchronous strengthening; The same path changes from weak correlation to strong correlation in a short period of time; A shift pattern not seen in historical data occurs for the first time; The conduction time of the influence path between variables is significantly shortened or mutated.

8. The multivariate trend anomaly detection method for home patients with heart failure according to claim 7 is characterized in that The determination of the pathological trend deviation satisfies at least two of the following conditions: Two or more causal paths shift in direction within the same cycle; The offset path does not appear in the historical trend graph; The intensity of trend transmission after the deviation exceeds the established risk threshold; The shift occurred along the typical high-risk HF variable pathway consisting of weight → respiratory rate → heart rate.

Images