A method, system, device, and storage medium for predicting the risk of a severe complication within the next 24 hours for a patient with a critical illness post-surgery

Abstract

The application provides a method, system, device and storage medium for predicting the risk of severe complications of a critically ill patient within 24 hours after surgery, the method comprising: acquiring preoperative stable period data and postoperative continuous multi-modal health data of the critically ill patient; preprocessing to obtain standardized time series data; constructing a state vector of the current postoperative state of the critically ill patient and a historical state sequence; constructing an individualized health baseline model based on the preoperative stable period data; comparing the state vector with the individualized health baseline model to obtain the deviation characteristics of the current postoperative state of the critically ill patient relative to the preoperative stable state; constructing a 24-hour risk evolution representation; and calculating the overall risk probability of the critically ill patient of developing severe complications within 24 hours after surgery. The application can realize early prediction of the risk of postoperative severe complications, and is more suitable for continuous monitoring, active review and early intervention of the high-risk window period of the critically ill patient after surgery.

Description

Technical Field

[0001] This invention relates to the field of medical and health monitoring and analysis technology, and in particular to a method, system, device and storage medium for predicting the risk of serious complications in critically ill patients within 24 hours after surgery. Background Technology

[0002] For cancer patients, patients undergoing major surgery, patients with severe cardiovascular and cerebrovascular diseases, organ transplant recipients, and patients with other serious illnesses, the early postoperative recovery period is typically a high-risk window for serious complications. During this stage, patients may experience worsening infections or inflammation, bleeding or circulatory instability, deteriorating respiratory function, or exacerbated hypoxia. Once these serious complications progress to a clearly defined clinical event stage, it often indicates disease progression, significantly increasing the difficulty of intervention and treatment costs.

[0003] Most existing postoperative monitoring protocols revolve around identifying current abnormalities. For example, they use fixed thresholds to determine if the heart rate is too fast, blood oxygen saturation is too low, or body temperature is too high, or they determine if an abnormal event has occurred based on deviations from current characteristics within a continuous time window. These protocols primarily address the problem of identifying current abnormalities, rather than predicting future risks.

[0004] However, serious complications typically do not occur suddenly at a single point in time, but rather evolve through a process of slight deviation, gradual deterioration, enhanced cross-indicator coupling, and eventual clinical exposure. For example, even if a patient has not yet reached a clear critical threshold at a certain point after surgery, but their resting heart rate continues to rise, blood oxygen levels slowly decrease, nighttime sleep fragmentation worsens, body temperature shows a gradual upward trend, and pain scores continue to rise, it is often difficult to identify the risk of serious complications in the future if judgment is based solely on whether the current abnormality has occurred.

[0005] Furthermore, there are significant differences among patients in their vital signs, diurnal rhythms, range of activity changes, and baseline subjective symptoms during their preoperative stable state. Relying solely on general thresholds or deviations from current characteristics for identification can easily lead to false positives and false negatives, and it is also difficult to provide a prospective risk assessment for critically ill patients during the postoperative critical window period.

[0006] Therefore, it is necessary to propose a new postoperative monitoring protocol that is not limited to identifying whether the patient is currently abnormal, but rather predicts the overall risk of serious complications in the next 24 hours based on the patient's preoperative individual baseline, current postoperative status, and recent evolution trend, thereby providing a basis for clinical early review, proactive intervention, and graded treatment. Summary of the Invention

[0007] To address the problems existing in the prior art, the purpose of this invention is to provide a method, system, device, and storage medium for predicting the risk of serious complications occurring in critically ill patients within 24 hours after surgery, for probabilistic prediction of the overall risk of serious complications occurring in critically ill patients within 24 hours after surgery.

[0008] One technical solution of the present invention is: providing a method for predicting the risk of serious complications in critically ill patients within 24 hours after surgery, comprising: Acquire preoperative stabilization data and postoperative continuous multimodal health data of critically ill patients; The preoperative stable period data and the postoperative continuous multimodal health data were preprocessed to obtain standardized time-series data; Based on multiple consecutive historical time windows prior to the current moment, current state features, trend evolution features, fluctuation evolution features, cross-index coupling evolution features, and symptom evolution features are extracted from the standardized time series data to construct the state vector of the current postoperative state of critically ill patients and the historical state sequence. An individualized health baseline model was constructed based on the preoperative stabilization period data. The state vector is compared with the individualized health baseline model to obtain the deviation characteristics of the current postoperative state of critically ill patients relative to their preoperative stable state. Based on the state vector, the historical state sequence, the deviation feature, the trend evolution feature, the fluctuation evolution feature, the cross-indicator coupling evolution feature, and the symptom evolution feature, a risk evolution representation for the next 24 hours is constructed. Based on the risk evolution representation for the next 24 hours, the overall risk probability of serious complications occurring in critically ill patients within the next 24 hours after surgery is calculated.

[0009] Furthermore, the preoperative stable period data and postoperative continuous multimodal health data of critically ill patients are obtained, including: the postoperative continuous multimodal health data includes vital sign data and subjective symptom data.

[0010] Furthermore, the preoperative stable period data and the postoperative continuous multimodal health data are preprocessed to obtain the standardized time-series data, including: The preprocessing includes time alignment, outlier suppression, noise filtering, missing value completion, and / or signal quality assessment.

[0011] Furthermore, based on multiple consecutive historical time windows prior to the current moment, current state features, trend evolution features, fluctuation evolution features, cross-index coupling evolution features, and symptom evolution features are extracted from the standardized time-series data to construct a state vector and historical state sequence of the current postoperative state of critically ill patients; including: The current state characteristics include statistical characteristics, trend characteristics, fluctuation characteristics, circadian rhythm characteristics, scene correction characteristics, and / or subjective symptom characteristics.

[0012] Furthermore, obtaining the cross-index coupling evolution characteristics includes the following steps: Construct a cross-index coupling matrix based on the standardized multimodal observation matrix within the current time window; Coupling deviation characteristics are obtained by comparing the cross-index coupling matrix in the current time window with the preoperative stable period reference coupling matrix; The cross-index coupling evolution characteristics are obtained based on the changing trends of the coupling deviation characteristics over multiple consecutive historical time windows.

[0013] Furthermore, an individualized health baseline model is constructed based on the preoperative stabilization period data; including: The individualized health baseline model is constructed based on monitoring data of the patient during a continuous predetermined period of preoperative stabilization. It is divided into scenarios according to resting state, activity state and sleep state, and multiple sub-baseline models are constructed according to daytime and nighttime periods. Each sub-baseline model includes a baseline mean vector, a baseline covariance matrix and a low-dimensional embedding center vector.

[0014] Furthermore, the state vector is compared with the individualized health baseline model to obtain the deviation characteristics of the current postoperative state of the critically ill patient relative to the preoperative stable state; including the following steps: Select the matching sub-baseline model based on the current physiological state and the current time period; The normalized deviation of the current postoperative condition from the preoperative stable baseline was calculated as follows: Dt=sqrt((St-μ(g,p))T·Σ(g,p)^(-1)·(St-μ(g,p))); where St represents the current postoperative state vector, μ(g,p) represents the baseline mean vector of the corresponding state-time sub-baseline model, Σ(g,p) represents the baseline covariance matrix of the corresponding state-time sub-baseline model, and Dt represents the normalized deviation distance in the deviation feature.

[0015] Furthermore, based on the aforementioned risk evolution representation for the next 24 hours, the overall risk probability of serious complications occurring in critically ill patients within the next 24 hours after surgery is calculated; including: The next 24 hours are divided into multiple forecast sub-intervals; Based on the risk evolution representation for the next 24 hours, the conditional risk rate of serious complications occurring within each prediction sub-interval is calculated; The conditional hazard rates of each prediction sub-interval are aggregated to obtain the overall risk probability P24h for the next 24 hours; the overall risk probability P24h satisfies: P24h = 1 - ∏(1-h(t+τ)); Where h(t+τ) represents the conditional risk rate of serious complications occurring in the τ-th prediction sub-interval in the future.

[0016] Another technical solution of the present invention is: providing a system for predicting the risk of serious complications in critically ill patients within 24 hours after surgery, comprising: The data acquisition module is used to acquire preoperative stable period data and postoperative continuous multimodal health data of critically ill patients; The data preprocessing module is used to preprocess the preoperative stable period data and the postoperative continuous multimodal health data to obtain standardized time-series data. The feature construction module extracts current state features, trend evolution features, fluctuation evolution features, cross-index coupling evolution features, and symptom evolution features from the standardized time series data based on multiple consecutive historical time windows prior to the current moment, and constructs the state vector of the current postoperative state of critically ill patients and the historical state sequence. The baseline modeling module constructs an individualized health baseline model based on the preoperative stable period data. The deviation analysis module is used to compare the state vector with the individualized health baseline model to obtain the deviation characteristics of the current postoperative state of the critically ill patient relative to the preoperative stable state. The risk evolution modeling module constructs a risk evolution representation for the next 24 hours based on the state vector, the historical state sequence, the deviation feature, the trend evolution feature, the fluctuation evolution feature, the cross-indicator coupling evolution feature, and the symptom evolution feature. The 24-hour risk prediction module is used to calculate the overall risk probability of serious complications occurring in critically ill patients within the next 24 hours after surgery, based on the 24-hour risk evolution representation.

[0017] Furthermore, the postoperative continuous multimodal health data includes vital sign data and subjective symptom data.

[0018] Furthermore, the data preprocessing module is used to preprocess the preoperative stable period data and the postoperative continuous multimodal health data. The preprocessing includes time alignment, outlier suppression, noise filtering, missing value completion, and / or signal quality assessment.

[0019] Furthermore, the feature construction module extracts current state features, trend evolution features, fluctuation evolution features, cross-indicator coupling evolution features, and symptom evolution features from the standardized time series data based on multiple consecutive historical time windows prior to the current moment. The current state characteristics include statistical characteristics, trend characteristics, fluctuation characteristics, circadian rhythm characteristics, scene correction characteristics, and subjective symptom characteristics.

[0020] Furthermore, obtaining the cross-index coupling evolution characteristics includes: The coupling matrix construction module constructs a cross-index coupling matrix based on the standardized multimodal observation matrix within the current time window; The coupling deviation module is used to obtain coupling deviation characteristics by comparing the difference between the cross-index coupling matrix in the current time window and the preoperative stable period reference coupling matrix; The cross-index coupling evolution feature acquisition module obtains the cross-index coupling evolution feature based on the changing trend of the coupling deviation feature in multiple consecutive historical time windows.

[0021] Furthermore, the baseline modeling module includes: The individualized health baseline model is constructed based on monitoring data of the patient during a continuous predetermined period of preoperative stabilization. It is divided into scenarios according to resting state, activity state and sleep state, and multiple sub-baseline models are constructed according to daytime and nighttime periods. Each sub-baseline model includes a baseline mean vector, a baseline covariance matrix and a low-dimensional embedding center vector.

[0022] Furthermore, the deviation analysis module includes: Select the matching sub-baseline model based on the current physiological state and the current time period; The normalized deviation of the current postoperative condition from the preoperative stable baseline was calculated as follows: Dt=sqrt((St-μ(g,p))T·Σ(g,p)^(-1)·(St-μ(g,p))); Where St represents the current postoperative state vector, μ(g,p) represents the baseline mean vector of the corresponding state-time sub-baseline model, Σ(g,p) represents the baseline covariance matrix of the corresponding state-time sub-baseline model, and Dt represents the normalized deviation distance in the deviation feature.

[0023] Furthermore, the risk prediction module for the next 24 hours includes: The next 24 hours are divided into multiple forecast sub-intervals; Based on the risk evolution representation for the next 24 hours, the conditional risk rate of serious complications occurring within each prediction sub-interval is calculated; The conditional hazard rates of each prediction sub-interval are aggregated to obtain the overall risk probability P24h for the next 24 hours; the overall risk probability P24h satisfies: P24h = 1 - ∏(1-h(t+τ)); Where h(t+τ) represents the conditional risk rate of serious complications occurring in the τ-th prediction sub-interval in the future.

[0024] The third technical solution of the present invention is: to provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-mentioned method for predicting the risk of serious complications in critically ill patients within 24 hours after surgery.

[0025] The fourth technical solution of the present invention is to provide a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the above-mentioned method for predicting the risk of serious complications in critically ill patients within 24 hours after surgery.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention shifts the monitoring target from "current anomaly identification" to "overall risk prediction of serious complications in the next 24 hours," providing an early warning before a clear clinical event occurs. It constructs an individualized health baseline based on preoperative stable period data of critically ill patients and combines this with the evolutionary trends of postoperative multimodal continuous data for prediction, reducing false positives and false negatives caused by simply using general thresholds or current anomaly identification. It considers not only the degree of deviation from the current state but also the rate of deterioration, recovery rate, coupling enhancement trend, symptom aggravation trend, and the degree of rhythm disruption, thus more accurately depicting the gradual evolutionary process before the formation of serious complications. This invention outputs the overall risk probability of serious complications occurring in the next 24 hours, enabling early prediction of postoperative serious complication risks and is more suitable for continuous monitoring, proactive review, and early intervention during the high-risk window period after surgery for critically ill patients. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the method for predicting the risk of serious complications in critically ill patients within 24 hours after surgery, as described in this invention. Figure 2 This is a structural block diagram of the system in this invention for predicting the risk of serious complications in critically ill patients within 24 hours after surgery; Figure 3 This is a diagram showing the internal structure of the computer device in this invention.

[0028] Figure label: 1. Data acquisition module; 2. Data preprocessing module; 3. Feature construction module; 4. Baseline modeling module; 5. Deviation analysis module; 6. Risk evolution modeling module; 7. Risk prediction module for the next 24 hours; 8. Risk output module. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. I. Please refer to Figure 1 , Figure 1 This paper discloses a method for predicting the risk of serious complications in critically ill patients within 24 hours after surgery, including the following steps: S100. Obtain preoperative stabilization data and postoperative continuous multimodal health data for critically ill patients; S200. Preprocess the preoperative stable period data and postoperative continuous multimodal health data to obtain standardized time series data; S300. Based on multiple consecutive historical time windows prior to the current moment, extract current state features, trend evolution features, fluctuation evolution features, cross-index coupling evolution features, and symptom evolution features from standardized time series data to construct the state vector of the current postoperative state of critically ill patients and the historical state sequence. S400. Construct an individualized health baseline model based on preoperative stable period data; S500. Compare the state vector with the individualized health baseline model to obtain the deviation characteristics of the current postoperative state of critically ill patients relative to their preoperative stable state; S600. Based on state vectors, historical state sequences, deviation features, trend evolution features, volatility evolution features, cross-indicator coupling evolution features, and symptom evolution features, a risk evolution representation for the next 24 hours is constructed. S700. Calculate the overall risk probability of serious complications occurring in critically ill patients within the next 24 hours after surgery, based on the risk evolution representation for the next 24 hours.

[0030] In this method, serious complications include worsening infection or inflammation, bleeding or circulatory instability, worsening respiratory function and / or exacerbation of hypoxia.

[0031] In step S100 of this method, for critically ill patients, baseline data is collected continuously for at least 3 days during the preoperative stable period, preferably for 3 to 14 days, to obtain preoperative stable period data for critically ill patients. Preoperative stable period data is used to characterize the individualized vital sign levels, rhythm characteristics, and baseline subjective symptoms of critically ill patients in a non-complication state. Vital sign data and subjective symptom data can be collected during both the preoperative stable period (preoperative stable period data) and the postoperative monitoring phase (postoperative continuous multimodal health data), as detailed below: 1. Heart rate data, in beats per minute; 2. Blood oxygen saturation data, in percentage; 3. Activity data, including at least one of the following: steps, activity intensity, and acceleration characteristics; 4. Sleep state data, including sleep onset time, light sleep duration, deep sleep duration, number of awakenings, and / or sleep fragmentation; 5. Body temperature data; 6. Blood pressure data, including systolic and diastolic blood pressure; 7. Subjective symptom data, including pain scores, mood scores, subjective descriptions of discomfort, and / or medication records.

[0032] During the postoperative monitoring phase, the above data are continuously recorded according to a unified timestamp to constitute continuous multimodal health data after surgery. It should be noted that vital sign data collected using wearable devices and subjective symptom data entered by patients with serious illnesses via a mobile app or mobile terminal can be used.

[0033] In step S200 of this method, the preoperative stable period data and postoperative continuous multimodal health data are preprocessed to obtain standardized time series data.

[0034] Preprocessing includes time alignment, outlier suppression, noise removal, missing value completion, and / or signal quality assessment. Outlier suppression includes hard thresholding based on physiologically feasible intervals and soft anomaly correction based on Hamper filtering, sliding median filtering, or outlier identification models. Noise removal includes Kalman filtering, wavelet denoising, and / or exponential smoothing. Missing value completion includes a hybrid approach based on temporal neighborhood interpolation, state-similar segment matching, and matrix completion.

[0035] The specific steps of preprocessing are as follows: For the collected raw time-series data, time synchronization and resampling are first performed to align data with different sampling frequencies to a unified time axis.

[0036] For vital signs such as heart rate, blood oxygen, body temperature, and blood pressure, hard threshold screening can be performed based on physiologically feasible ranges. For example, when the heart rate is less than 30 beats / minute or greater than 220 beats / minute, it can be marked as a suspected abnormal sampling point; when the blood oxygen saturation is less than 50%, it can be marked as a suspected invalid point.

[0037] Noise is suppressed and local outliers are corrected by using methods such as Hamper filtering, sliding median filtering, Kalman filtering, wavelet denoising, or exponential smoothing.

[0038] For missing time points, linear interpolation, spline interpolation, local neighborhood matching, similar segment-based completion, or matrix completion can be used to repair them, depending on the data type.

[0039] For data such as activity levels and sleep status, signal continuity, effective coverage, and sampling stability can be further calculated to form a signal quality score Qt. This signal quality score can serve as one of the inputs for subsequent risk evolution modeling and risk probability correction.

[0040] In step S300 of this method, based on multiple consecutive historical time windows prior to the current moment, current state features, trend evolution features, fluctuation evolution features, cross-index coupling evolution features, and symptom evolution features are extracted from standardized time series data to construct the state vector of the current postoperative state of critically ill patients and the historical state sequence.

[0041] Current state characteristics include statistical features, trend features, fluctuation features, circadian rhythm features, scene-corrected features, and / or subjective symptom features. Statistical features include window mean, standard deviation, median, range, and / or quantiles; trend features include slope of change, first-order difference mean, and / or acceleration of change; fluctuation features include coefficient of variation, stability index, and / or recovery rate; scene-corrected features include activity-corrected heart rate deviation, blood oxygen fluctuation amplitude during sleep, and / or body temperature deviation at rest.

[0042] Trend evolution characteristics include recent slope changes, rate of deterioration, and / or rate of recovery; fluctuation evolution characteristics include the rate of change in stability and / or the duration of anomalous fluctuations. Cross-indicator coupling evolution characteristics include the strength of coupling deviation and / or the trend of coupling enhancement. Symptom evolution characteristics include trends in pain changes, trends in mood deterioration, and / or the cumulative intensity of subjective discomfort.

[0043] In step S300 of this method, obtaining the cross-index coupling evolution characteristics includes the following steps: Construct a cross-index coupling matrix based on the standardized multimodal observation matrix within the current time window; Coupling deviation characteristics are obtained by comparing the difference between the current time window cross-index coupling matrix and the preoperative stable period reference coupling matrix; Based on the changing trends of coupling deviation characteristics in multiple consecutive historical time windows, cross-index coupling evolution characteristics are obtained.

[0044] Based on the above steps, the specific algorithm for obtaining cross-index coupled evolutionary features is as follows: In a preferred embodiment, a cross-index coupling matrix Ct can be constructed based on the standardized matrix X't: Ct = (1 / (n-1))·(X't)T·X't; Ct is used to characterize the synergistic changes among different vital sign indicators and symptom indicators. By comparing the difference between the current time window coupling matrix Ct and the preoperative stable period reference coupling matrix, coupling deviation characteristics can be obtained.

[0045] Furthermore, by analyzing the changing trends of coupling deviation features across multiple consecutive historical time windows, coupling enhancement trend features can be obtained. These features can reflect the evolutionary process in critically ill patients after surgery, where a single deviation gradually develops into a synergistic deterioration of multiple indicators.

[0046] In step S300 of this method, the step of extracting the current state features from the standardized time series data is as follows: Define multiple sliding time windows, such as at least one of 5 minutes, 15 minutes, 60 minutes, and 24 hours. For any current time t, construct an observation matrix Xt using multimodal data from multiple consecutive historical time windows prior to the current time.

[0047] Let Xt∈R^(n×m), where n represents the number of sampling points within the window and m represents the number of monitoring indicators. Normalizing Xt yields the standardized matrix X't. Based on the standardized matrix X't, the following features are extracted: Statistical features, including mean, median, maximum, minimum, standard deviation, range, and quantiles. Trend features, including first-order difference mean, regression slope, acceleration of change, and trend continuity. Fluctuation features, including coefficient of variation, stability index, duration of fluctuation, and recovery speed. Circadian rhythm features, including nighttime resting deviation, diurnal amplitude difference, and similarity to the preoperative rhythm template. Scene-corrected features, including activity-corrected heart rate deviation, blood oxygen fluctuation amplitude during sleep, and body temperature deviation during rest. Subjective symptom features, including mean pain score, rate of pain change, emotional fluctuation, and encoding of subjective discomfort intensity.

[0048] After concatenating the above features, a state vector St (current state features) is formed for the current postoperative state of critically ill patients.

[0049] In step S400 of this method, an individualized health baseline model is constructed based on preoperative stable period data.

[0050] The individualized health baseline model is constructed based on monitoring data of critically ill patients during a continuous predetermined period of time (preferably 3 to 14 consecutive days) before surgery. It is divided into scenarios according to resting state, activity state and sleep state, and multiple sub-baseline models are constructed according to daytime and nighttime periods. Each sub-baseline model includes a baseline mean vector, a baseline covariance matrix and a low-dimensional embedding center vector.

[0051] For each sub-baseline model, the corresponding baseline mean vector μ(g,p) and baseline covariance matrix Σ(g,p) can be calculated. The baseline mean vector represents the typical health level of critically ill patients in the corresponding state, and the baseline covariance matrix represents the normal combined fluctuation range of each indicator.

[0052] To further enhance the expressive power of the personalized health baseline model, principal component analysis, autoencoder, or other dimensionality reduction mapping models can be used to map the high-dimensional feature vectors of the preoperative stable period to a low-dimensional latent space, thereby obtaining the low-dimensional embedding center vector z(g,p).

[0053] In a preferred embodiment, the personalized health baseline model can also save the reliability weights of different indicators, the reference threshold bands under different states, and the individual sensitivity factors of patients with serious illnesses for subsequent deviation analysis and risk evolution modeling.

[0054] In step S500 of this method, the state vector is compared with an individualized health baseline model to obtain the deviation characteristics of the critically ill patient's current postoperative state relative to the preoperative stable state. The deviation characteristics include at least the normalized deviation distance, state shift direction, coupling deviation strength, and the degree of rhythm disruption relative to the preoperative rhythm template. This includes the following steps: Select a matching sub-baseline model based on the current physiological state and the current time period; The normalized deviation distance of the current postoperative state relative to the preoperative stable baseline is calculated as follows: Dt=sqrt((St-μ(g,p))T·Σ(g,p)^(-1)·(St-μ(g,p))); where St represents the current postoperative state vector, μ(g,p) represents the baseline mean vector of the corresponding state-time sub-baseline model, Σ(g,p) represents the baseline covariance matrix of the corresponding state-time sub-baseline model, and Dt represents the normalized deviation distance in the deviation feature.

[0055] In step S600 of this method, a risk evolution representation for the next 24 hours is constructed based on state vectors, historical state sequences, deviation features, trend evolution features, fluctuation evolution features, cross-indicator coupling evolution features, and symptom evolution features. It should be noted that the method does not directly determine that a critically ill patient is "currently abnormal" based on current deviation features, but rather further models the recent continuous state sequence of the critically ill patient to predict the overall risk of serious complications within the next 24 hours.

[0056] The risk evolution representation for the next 24 hours is obtained through a future risk prediction model. This model can be a gated recurrent network, a temporal convolutional network, a temporal attention network, a survival analysis model, or a combination thereof. The future risk prediction model is used to model the deterioration rate, recovery rate, intensity of persistent deviation, degree of circadian rhythm disruption, and symptom exacerbation trend reflected in historical state sequences to generate a risk evolution representation for the next 24 hours.

[0057] The steps to construct a risk evolution representation for the next 24 hours are as follows.

[0058] The historical state sequence is constructed using the state vectors of K consecutive time windows prior to the current time t: Ut = [St-K+1, St-K+2, ..., St]. Based on the historical state sequence Ut, the following evolutionary features can be extracted: 1. Deterioration rate characteristics, used to characterize the rate of increase of normalized deviation distance, coupling deviation strength, or symptom intensity; 2. Recovery speed characteristics, used to characterize the speed at which vital signs and symptom indicators regress to preoperative baseline; 3. Persistent deviation intensity feature, used to characterize the cumulative degree of abnormal deviation within a continuous historical time window; 4. Characteristics of the degree of rhythm disruption, used to measure the degree of disruption of the current diurnal rhythm of critically ill patients relative to the preoperative rhythm template; 5. Symptom exacerbation trend characteristics are used to characterize the continuous worsening trend of pain, mood, and subjective discomfort in the recent window.

[0059] The current state vector St, the historical state sequence Ut, the deviation features, and the aforementioned evolutionary features are input into the future risk prediction model to obtain the risk evolution representation H(t,24h) for the next 24 hours. H(t,24h) is used for prospective information on the evolution of the current state of critically ill patients toward the risk of serious complications in the next 24 hours.

[0060] In step S700 of this method, the overall risk probability of serious complications occurring in critically ill patients within the next 24 hours after surgery is calculated based on the risk evolution representation for the next 24 hours; including: Divide the next 24 hours into multiple prediction sub-intervals; for example, every 1 hour, every 2 hours, every 4 hours, or other preset durations can be used as a prediction sub-interval.

[0061] Based on the risk evolution representation for the next 24 hours, the conditional risk rate of serious complications occurring within each prediction sub-interval is calculated; The conditional hazard rates of each prediction sub-interval are aggregated to obtain the overall risk probability P24h for the next 24 hours; the overall risk probability P24h satisfies: P24h = 1 - ∏(1-h(t+τ)); Where h(t+τ) represents the conditional risk rate of serious complications occurring in the τ-th prediction sub-interval in the future.

[0062] This method also includes step S800, which outputs the risk level for the next 24 hours and the corresponding early warning or intervention prompts based on the overall risk probability.

[0063] The risk level for the next 24 hours will include at least four levels: low risk, medium risk, high risk, and risk requiring key intervention. Specifically, a low risk level is defined as the overall risk probability being below the first threshold; a medium risk level is defined as the overall risk probability being above or equal to the first threshold but below the second threshold; a high risk level is defined as the overall risk probability being above or equal to the second threshold but below the third threshold; and a risk requiring key intervention is defined as the overall risk probability being above or equal to the third threshold, or, although the overall risk probability does not reach the third threshold, it simultaneously meets the conditions of continuous deterioration and abnormal combination of core indicators. The first, second, and third thresholds are in an increasing order. The first, second, and third thresholds can be set based on the patient's physical condition or data obtained from clinical trials.

[0064] Conditions of continued deterioration may include at least one of the following: continuously increasing normalized deviation, continuously decreasing blood oxygen, continuously increasing resting heart rate, continuously increasing body temperature, or continuously worsening symptom score; abnormal conditions of core indicator combination may include: increased heart rate accompanied by decreased blood oxygen during low activity, or increased body temperature accompanied by increased pain during rest at night, etc.

[0065] When the system outputs a medium risk, it can push a retest and enhanced observation prompt to the critically ill patient's end; when the system outputs a high risk, it can push a suggestion to the medical staff to review infection indicators, oxygenation status, circulatory status, or bleeding-related indicators; when the system outputs a key intervention risk, it can simultaneously trigger a high-priority prompt to support early intervention.

[0066] In this method, when a critically ill patient remains in a stable recovery state for a continuously preset period after surgery and the data quality score is consistently higher than the update threshold, the parameters of the individualized health baseline model or future risk prediction model are adaptively updated in a low-weight manner to adapt to the physiological reconstruction during the long-term recovery process of critically ill patients after surgery.

[0067] For critically ill patients who continue to recover well post-surgery, new steady-state differences may exist between the pre-operative baseline and the long-term stable state after surgery. To avoid continuously misclassifying new stable states as high-risk in the long-term monitoring phase, this invention performs low-weight adaptive updates when the following conditions are met: 1. Critically ill patients remain in a stable recovery state for a preset duration; 2. The overall risk probability remains below the updated threshold for the next 24 hours; 3. No emergency combinations of core indicators showed abnormalities; 4. The data quality score consistently exceeds the preset value.

[0068] When the above conditions are met, recent stable window features can be incorporated into the current individualized health baseline model with a lower update weight, or the parameters of the future risk prediction model can be gently updated to improve the adaptability in the long-term follow-up phase.

[0069] Compared with existing technologies, this invention shifts the monitoring target from "current anomaly identification" to "overall risk prediction of serious complications in the next 24 hours," providing an early warning before a clear clinical event occurs. By constructing an individualized health baseline based on preoperative stable period data of critically ill patients and combining it with the evolutionary trends of postoperative multimodal continuous data for prediction, it can reduce false alarms and missed alarms caused by simply using general thresholds or current anomaly identification. It considers not only the degree of deviation from the current state but also the rate of deterioration, recovery rate, coupling enhancement trend, symptom aggravation trend, and the degree of rhythm disruption, thus more accurately depicting the gradual evolutionary process before the formation of serious complications. This invention outputs the overall risk probability and risk level of serious complications occurring in the next 24 hours, enabling early prediction of postoperative serious complication risks and is more suitable for continuous monitoring, proactive review, and early intervention during the high-risk window period after surgery for critically ill patients.

[0070] III. Please refer to Figure 2 , Figure 2 What is revealed is a system for predicting the risk of serious complications in critically ill patients within the next 24 hours after surgery, including: Data acquisition module 1 is used to acquire preoperative stable period data and postoperative continuous multimodal health data of critically ill patients; Data preprocessing module 2 is used to preprocess preoperative stable period data and postoperative continuous multimodal health data to obtain standardized time series data; Feature construction module 3 extracts current state features, trend evolution features, fluctuation evolution features, cross-index coupling evolution features, and symptom evolution features from standardized time series data based on multiple consecutive historical time windows prior to the current moment, and constructs the state vector of the current postoperative state of critically ill patients and the historical state sequence. Baseline modeling module 4: Constructs an individualized health baseline model based on preoperative stable period data; Deviation analysis module 5 is used to compare the state vector with the individualized health baseline model to obtain the deviation characteristics of the current postoperative state of critically ill patients relative to their preoperative stable state. Risk evolution modeling module 6 constructs a risk evolution representation for the next 24 hours based on state vectors, historical state sequences, deviation features, trend evolution features, volatility evolution features, cross-indicator coupling evolution features, and symptom evolution features. The 24-hour risk prediction module 7 is used to calculate the overall risk probability of serious complications occurring in critically ill patients within the next 24 hours after surgery, based on the risk evolution representation of the next 24 hours.

[0071] In the embodiments, serious complications include worsening infection or inflammation, bleeding or circulatory instability, worsening respiratory function and / or exacerbation of hypoxia.

[0072] In this embodiment, postoperative continuous multimodal health data includes vital sign data and subjective symptom data.

[0073] For critically ill patients, baseline data should be collected continuously for at least 3 days during the preoperative stabilization period, preferably 3 to 14 days, to obtain preoperative stabilization data. Preoperative stabilization data is used to characterize the individualized vital sign levels, rhythm characteristics, and baseline subjective symptoms of critically ill patients in a non-complication state. Vital sign data and subjective symptom data can be collected during both the preoperative stabilization period (preoperative stabilization data) and the postoperative monitoring phase (postoperative continuous multimodal health data), as detailed below: 1. Heart rate data, in beats per minute; 2. Blood oxygen saturation data, in percentage; 3. Activity data, including at least one of the following: steps, activity intensity, and acceleration characteristics; 4. Sleep state data, including sleep onset time, light sleep duration, deep sleep duration, number of awakenings, and / or sleep fragmentation; 5. Body temperature data; 6. Blood pressure data, including systolic and diastolic blood pressure; 7. Subjective symptom data, including pain scores, mood scores, subjective descriptions of discomfort, and / or medication records.

[0074] During the postoperative monitoring phase, the above data are continuously recorded according to a unified timestamp to constitute continuous multimodal health data after surgery. It should be noted that vital sign data collected using wearable devices and subjective symptom data entered by patients with serious illnesses via a mobile app or mobile terminal can be used.

[0075] In this embodiment, the data preprocessing module 2 is used to preprocess the preoperative stable period data and the postoperative continuous multimodal health data to obtain standardized time series data. Preprocessing includes time alignment, outlier suppression, noise removal, missing value completion, and / or signal quality assessment. Outlier suppression includes hard thresholding based on physiologically feasible intervals and soft anomaly correction based on Hamper filtering, sliding median filtering, or outlier identification models. Noise removal includes Kalman filtering, wavelet denoising, and / or exponential smoothing. Missing value completion includes a hybrid approach based on temporal neighborhood interpolation, state-similar segment matching, and matrix completion.

[0076] The specific steps of preprocessing are as follows: For the collected raw time-series data, time synchronization and resampling are first performed to align data with different sampling frequencies to a unified time axis.

[0077] For vital signs such as heart rate, blood oxygen, body temperature, and blood pressure, hard threshold screening can be performed based on physiologically feasible ranges. For example, when the heart rate is less than 30 beats / minute or greater than 220 beats / minute, it can be marked as a suspected abnormal sampling point; when the blood oxygen saturation is less than 50%, it can be marked as a suspected invalid point.

[0078] Noise is suppressed and local outliers are corrected by using methods such as Hamper filtering, sliding median filtering, Kalman filtering, wavelet denoising, or exponential smoothing.

[0079] For missing time points, linear interpolation, spline interpolation, local neighborhood matching, similar segment-based completion, or matrix completion can be used to repair them, depending on the data type.

[0080] For data such as activity levels and sleep status, signal continuity, effective coverage, and sampling stability can be further calculated to form a signal quality score Qt. This signal quality score can serve as one of the inputs for subsequent risk evolution modeling and risk probability correction.

[0081] In this embodiment, feature construction module 3 extracts current state features, trend evolution features, fluctuation evolution features, cross-index coupling evolution features, and symptom evolution features from standardized time series data based on multiple consecutive historical time windows prior to the current moment, and constructs the state vector of the current postoperative state of the critically ill patient and the historical state sequence.

[0082] Current state characteristics include statistical features, trend features, fluctuation features, circadian rhythm features, scene-corrected features, and / or subjective symptom features. Statistical features include window mean, standard deviation, median, range, and / or quantiles; trend features include slope of change, first-order difference mean, and / or acceleration of change; fluctuation features include coefficient of variation, stability index, and / or recovery rate; scene-corrected features include activity-corrected heart rate deviation, blood oxygen fluctuation amplitude during sleep, and / or body temperature deviation at rest.

[0083] Trend evolution characteristics include recent slope changes, rate of deterioration, and / or rate of recovery; fluctuation evolution characteristics include the rate of change in stability and / or the duration of anomalous fluctuations. Cross-indicator coupling evolution characteristics include the strength of coupling deviation and / or the trend of coupling enhancement. Symptom evolution characteristics include trends in pain changes, trends in mood deterioration, and / or the cumulative intensity of subjective discomfort.

[0084] In feature construction module 3, the cross-index coupled evolution features are obtained as follows: The coupling matrix construction module constructs a cross-index coupling matrix based on the standardized multimodal observation matrix within the current time window; The coupling deviation module is used to obtain coupling deviation characteristics by comparing the difference between the cross-index coupling matrix of the current time window and the reference coupling matrix of the preoperative stable period; The cross-index coupling evolution feature acquisition module obtains cross-index coupling evolution features based on the changing trends of coupling deviation features in multiple consecutive historical time windows.

[0085] Based on the above, the specific algorithm for obtaining cross-index coupling evolutionary characteristics is as follows: In a preferred embodiment, a cross-index coupling matrix Ct can be constructed based on the standardized matrix X't: Ct = (1 / (n-1))·(X't)T·X't; Ct is used to characterize the synergistic changes among different vital sign indicators and symptom indicators. By comparing the difference between the current time window coupling matrix Ct and the preoperative stable period reference coupling matrix, coupling deviation characteristics can be obtained.

[0086] Furthermore, by analyzing the changing trends of coupling deviation features across multiple consecutive historical time windows, coupling enhancement trend features can be obtained. These features can reflect the evolutionary process in critically ill patients after surgery, where a single deviation gradually develops into a synergistic deterioration of multiple indicators.

[0087] In this embodiment, the steps for extracting the current state features from standardized time-series data are as follows: Define multiple sliding time windows, such as at least one of 5 minutes, 15 minutes, 60 minutes, and 24 hours. For any current time t, construct an observation matrix Xt using multimodal data from multiple consecutive historical time windows prior to the current time.

[0088] Let Xt∈R^(n×m), where n represents the number of sampling points within the window and m represents the number of monitoring indicators. Normalizing Xt yields the standardized matrix X't. Based on the standardized matrix X't, the following features are extracted: Statistical features, including mean, median, maximum, minimum, standard deviation, range, and quantiles. Trend features, including first-order difference mean, regression slope, acceleration of change, and trend continuity. Fluctuation features, including coefficient of variation, stability index, duration of fluctuation, and recovery speed. Circadian rhythm features, including nighttime resting deviation, diurnal amplitude difference, and similarity to the preoperative rhythm template. Scene-corrected features, including activity-corrected heart rate deviation, blood oxygen fluctuation amplitude during sleep, and body temperature deviation during rest. Subjective symptom features, including mean pain score, rate of pain change, emotional fluctuation, and encoding of subjective discomfort intensity.

[0089] After concatenating the above features, a state vector St (current state features) is formed for the current postoperative state of critically ill patients.

[0090] In this embodiment, the baseline modeling module 4 constructs an individualized health baseline model based on preoperative stable period data.

[0091] The individualized health baseline model is constructed based on monitoring data of critically ill patients during a continuous predetermined period of time (preferably 3 to 14 consecutive days) before surgery. It is divided into scenarios according to resting state, activity state and sleep state, and multiple sub-baseline models are constructed according to daytime and nighttime periods. Each sub-baseline model includes a baseline mean vector, a baseline covariance matrix and a low-dimensional embedding center vector.

[0092] For each sub-baseline model, the corresponding baseline mean vector μ(g,p) and baseline covariance matrix Σ(g,p) can be calculated. The baseline mean vector represents the typical health level of critically ill patients in the corresponding state, and the baseline covariance matrix represents the normal combined fluctuation range of each indicator.

[0093] To further enhance the expressive power of the personalized health baseline model, principal component analysis, autoencoder, or other dimensionality reduction mapping models can be used to map the high-dimensional feature vectors of the preoperative stable period to a low-dimensional latent space, thereby obtaining the low-dimensional embedding center vector z(g,p).

[0094] In a preferred embodiment, the personalized health baseline model can also save the reliability weights of different indicators, the reference threshold bands under different states, and the individual sensitivity factors of patients with serious illnesses for subsequent deviation analysis and risk evolution modeling.

[0095] In this embodiment, the deviation analysis module 5 is used to compare the state vector with the individualized health baseline model to obtain the deviation characteristics of the current postoperative state of the critically ill patient relative to the preoperative stable state. The deviation characteristics include at least the normalized deviation distance, state shift direction, coupling deviation strength, and the degree of rhythm disruption relative to the preoperative rhythm template. The deviation analysis module 5 includes: Select a matching sub-baseline model based on the current physiological state and the current time period; The normalized deviation distance of the current postoperative state relative to the preoperative stable baseline is calculated as follows: Dt=sqrt((St-μ(g,p))T·Σ(g,p)^(-1)·(St-μ(g,p))); where St represents the current postoperative state vector, μ(g,p) represents the baseline mean vector of the corresponding state-time sub-baseline model, Σ(g,p) represents the baseline covariance matrix of the corresponding state-time sub-baseline model, and Dt represents the normalized deviation distance in the deviation feature.

[0096] In this embodiment, the risk evolution modeling module 6 constructs a risk evolution representation for the next 24 hours based on state vectors, historical state sequences, deviation features, trend evolution features, fluctuation evolution features, cross-indicator coupling evolution features, and symptom evolution features. It should be noted that the system does not directly determine that a critically ill patient is "currently abnormal" based on current deviation features, but rather further models the recent continuous state sequence of the critically ill patient to predict the overall risk of serious complications within the next 24 hours.

[0097] The risk evolution representation for the next 24 hours is obtained through a future risk prediction model. This model can be a gated recurrent network, a temporal convolutional network, a temporal attention network, a survival analysis model, or a combination thereof. The future risk prediction model is used to model the deterioration rate, recovery rate, intensity of persistent deviation, degree of circadian rhythm disruption, and symptom exacerbation trend reflected in historical state sequences to generate a risk evolution representation for the next 24 hours.

[0098] The steps to construct a risk evolution representation for the next 24 hours are as follows.

[0099] The historical state sequence is constructed using the state vectors of K consecutive time windows prior to the current time t: Ut = [St-K+1, St-K+2, ..., St]. Based on the historical state sequence Ut, the following evolutionary features can be extracted: 1. Deterioration rate characteristics, used to characterize the rate of increase of normalized deviation distance, coupling deviation strength, or symptom intensity; 2. Recovery speed characteristics, used to characterize the speed at which vital signs and symptom indicators regress to preoperative baseline; 3. Persistent deviation intensity feature, used to characterize the cumulative degree of abnormal deviation within a continuous historical time window; 4. Characteristics of the degree of rhythm disruption, used to measure the degree of disruption of the current diurnal rhythm of critically ill patients relative to the preoperative rhythm template; 5. Symptom exacerbation trend characteristics are used to characterize the continuous worsening trend of pain, mood, and subjective discomfort in the recent window.

[0100] The current state vector St, the historical state sequence Ut, the deviation features, and the aforementioned evolutionary features are input into the future risk prediction model to obtain the risk evolution representation H(t,24h) for the next 24 hours. H(t,24h) is used for prospective information on the evolution of the current state of critically ill patients toward the risk of serious complications in the next 24 hours.

[0101] In this embodiment, the 24-hour risk prediction module 7 is used to calculate the overall risk probability of a critically ill patient developing serious complications within the next 24 hours after surgery, based on the 24-hour risk evolution representation. This includes: Divide the next 24 hours into multiple prediction sub-intervals; for example, every 1 hour, every 2 hours, every 4 hours, or other preset durations can be used as a prediction sub-interval.

[0102] Based on the risk evolution representation for the next 24 hours, the conditional risk rate of serious complications occurring within each prediction sub-interval is calculated; The conditional hazard rates of each prediction sub-interval are aggregated to obtain the overall risk probability P24h for the next 24 hours; the overall risk probability P24h satisfies: P24h = 1 - ∏(1-h(t+τ)); Where h(t+τ) represents the conditional risk rate of serious complications occurring in the τ-th prediction sub-interval in the future.

[0103] In this embodiment, a risk output module 8 is also included, which is used to output the risk level for the next 24 hours and the corresponding early warning or intervention prompts based on the overall risk probability.

[0104] The risk level for the next 24 hours will include at least four levels: low risk, medium risk, high risk, and risk requiring key intervention. Specifically, a low risk level is defined as the overall risk probability being below the first threshold; a medium risk level is defined as the overall risk probability being above or equal to the first threshold but below the second threshold; a high risk level is defined as the overall risk probability being above or equal to the second threshold but below the third threshold; and a risk requiring key intervention is defined as the overall risk probability being above or equal to the third threshold, or, although the overall risk probability does not reach the third threshold, it simultaneously meets the conditions of continuous deterioration and abnormal combination of core indicators. The first, second, and third thresholds are in an increasing order. The first, second, and third thresholds can be set based on the patient's physical condition or data obtained from clinical trials.

[0105] Conditions of continued deterioration may include at least one of the following: continuously increasing normalized deviation, continuously decreasing blood oxygen, continuously increasing resting heart rate, continuously increasing body temperature, or continuously worsening symptom score; abnormal conditions of core indicator combination may include: increased heart rate accompanied by decreased blood oxygen during low activity, or increased body temperature accompanied by increased pain during rest at night, etc.

[0106] When the system outputs a medium risk, it can push a retest and enhanced observation prompt to the critically ill patient's end; when the system outputs a high risk, it can push a suggestion to the medical staff to review infection indicators, oxygenation status, circulatory status, or bleeding-related indicators; when the system outputs a key intervention risk, it can simultaneously trigger a high-priority prompt to support early intervention.

[0107] In this embodiment, when a critically ill patient remains in a stable recovery state for a continuously preset period after surgery and the data quality score is consistently higher than the update threshold, the parameters of the individualized health baseline model or future risk prediction model are adaptively updated in a low-weight manner to adapt to the physiological reconstruction during the long-term recovery process of the critically ill patient after surgery.

[0108] For critically ill patients who continue to recover well post-surgery, new steady-state differences may exist between the pre-operative baseline and the long-term stable state after surgery. To avoid continuously misclassifying new stable states as high-risk in the long-term monitoring phase, this invention performs low-weight adaptive updates when the following conditions are met: 1. Critically ill patients remain in a stable recovery state for a preset duration; 2. The overall risk probability remains below the updated threshold for the next 24 hours; 3. No emergency combinations of core indicators showed abnormalities; 4. The data quality score consistently exceeds the preset value.

[0109] When the above conditions are met, recent stable window features can be incorporated into the current individualized health baseline model with a lower update weight, or the parameters of the future risk prediction model can be gently updated to improve the adaptability in the long-term follow-up phase.

[0110] The various modules in the aforementioned system for predicting the risk of serious complications in critically ill patients within the next 24 hours after surgery can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0111] III. To better understand the technical content of this invention, taking a postoperative patient with a serious illness as an example, the system establishes an individualized health baseline 7 days before surgery. At rest overnight before surgery, the patient's average heart rate was 68 beats / minute, blood oxygen saturation was 98%, body temperature was 36.6 degrees Celsius, sleep was stable, pain score was 1 to 2, and mood score was stable.

[0112] On the night of the third postoperative day, the system detected the following changes in the patient during multiple time windows under low activity and rest conditions: resting heart rate gradually increased from 84 beats / minute to 96 beats / minute, blood oxygen saturation slowly decreased from 97% to 94%, body temperature gradually increased from 37.0 degrees Celsius to 37.8 degrees Celsius, sleep fragmentation significantly increased, and the patient recorded a pain score of 3 to 6 points in the mini-program, accompanied by descriptions of anxiety and chest tightness.

[0113] At this point, the patient's various indicators have not yet all reached the preset critical threshold. If only the current abnormality identification scheme is used, it may not be sufficient to determine a clear serious abnormal event. However, this invention first constructs the current state vector St and compares it with the patient's preoperative nighttime resting sub-baseline model, obtaining a high normalized deviation distance Dt and obvious rhythm disruption characteristics. Subsequently, by combining the deterioration rate, continuous deviation intensity, coupling enhancement trend, and symptom aggravation trend in the historical state sequence, a risk evolution representation H(t,24h) for the next 24 hours is constructed.

[0114] Based on this, the system further calculates the conditional hazard rate of serious complications occurring in multiple predicted sub-intervals in the future, and aggregates them to obtain the overall risk probability P24h for the next 24 hours. For example, the system can output P24h as 0.72, corresponding to a high-risk level, and push a prompt to medical staff: "The overall risk of serious complications has increased in the next 24 hours. It is recommended to review infection indicators, oxygenation status, and circulatory status."

[0115] If the patient does indeed experience elevated infection markers and deterioration of respiratory function within the following 12 to 18 hours, it indicates that the present invention provides effective prospective risk prediction before the formation of serious complications, thereby gaining time for early clinical intervention.

[0116] As can be seen from the above embodiments, the present invention does not only focus on judging whether "the current situation is abnormal", but rather on the patient's individualized preoperative baseline, current postoperative status and recent evolution trajectory to make a probabilistic prediction of the overall risk of serious complications in the next 24 hours. Therefore, it is more suitable for continuous monitoring and proactive risk management of high-risk windows after surgery for critically ill patients.

[0117] IV. A computer device is provided, which can be a server or a terminal, and its internal structure diagram can be as follows. Figure 3 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and databases. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media to run. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When executed by a processor, this computer program enables a method for predicting the risk of serious complications in critically ill patients within 24 hours after surgery. Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements. V. A computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the steps in the above-described method embodiments. Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these. The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for predicting the risk of serious complications in critically ill patients within 24 hours after surgery, characterized in that, include: Acquire preoperative stabilization data and postoperative continuous multimodal health data of critically ill patients; The preoperative stable period data and the postoperative continuous multimodal health data were preprocessed to obtain standardized time-series data; Based on multiple consecutive historical time windows prior to the current moment, current state features, trend evolution features, fluctuation evolution features, cross-index coupling evolution features, and symptom evolution features are extracted from the standardized time series data to construct the state vector of the current postoperative state of critically ill patients and the historical state sequence. An individualized health baseline model was constructed based on the preoperative stabilization period data. The state vector is compared with the individualized health baseline model to obtain the deviation characteristics of the current postoperative state of critically ill patients relative to their preoperative stable state. Based on the state vector, the historical state sequence, the deviation feature, the trend evolution feature, the fluctuation evolution feature, the cross-indicator coupling evolution feature, and the symptom evolution feature, a risk evolution representation for the next 24 hours is constructed. Based on the risk evolution representation for the next 24 hours, the overall risk probability of serious complications occurring in critically ill patients within the next 24 hours after surgery is calculated.

2. The method according to claim 1, characterized in that, Acquire the preoperative stable period data and the postoperative continuous multimodal health data of critically ill patients, including: the postoperative continuous multimodal health data includes vital sign data and subjective symptom data.

3. The method according to claim 1, characterized in that, The preoperative stable period data and the postoperative continuous multimodal health data are preprocessed to obtain the standardized time-series data, including: The preprocessing includes time alignment, outlier suppression, noise filtering, missing value completion, and / or signal quality assessment.

4. The method according to claim 1, characterized in that, Based on multiple consecutive historical time windows prior to the current moment, current state features, trend evolution features, fluctuation evolution features, cross-index coupling evolution features, and symptom evolution features are extracted from the standardized time-series data to construct a state vector and historical state sequence of the current postoperative state of critically ill patients; including: The current state characteristics include statistical characteristics, trend characteristics, fluctuation characteristics, circadian rhythm characteristics, scene correction characteristics, and / or subjective symptom characteristics.

5. The method according to claim 4, characterized in that, Obtaining the cross-index coupling evolution features includes the following steps: Construct a cross-index coupling matrix based on the standardized multimodal observation matrix within the current time window; Coupling deviation characteristics are obtained by comparing the cross-index coupling matrix in the current time window with the preoperative stable period reference coupling matrix; The cross-index coupling evolution characteristics are obtained based on the changing trends of the coupling deviation characteristics over multiple consecutive historical time windows.

6. The method according to claim 1, characterized in that, An individualized health baseline model is constructed based on the preoperative stabilization period data; including: The individualized health baseline model is constructed based on monitoring data of the patient during a continuous predetermined period of preoperative stabilization. It is divided into scenarios according to resting state, activity state and sleep state, and multiple sub-baseline models are constructed according to daytime and nighttime periods. Each sub-baseline model includes a baseline mean vector, a baseline covariance matrix and a low-dimensional embedding center vector.

7. The method according to claim 6, characterized in that, The state vector is compared with the individualized health baseline model to obtain the deviation characteristics of the current postoperative state of the critically ill patient relative to the preoperative stable state; including the following steps: Select the matching sub-baseline model based on the current physiological state and the current time period; The normalized deviation of the current postoperative condition from the preoperative stable baseline was calculated as follows: Dt=sqrt((St-μ(g,p))T·Σ(g,p)^(-1)·(St-μ(g,p))); where St represents the current postoperative state vector, μ(g,p) represents the baseline mean vector of the corresponding state-time sub-baseline model, Σ(g,p) represents the baseline covariance matrix of the corresponding state-time sub-baseline model, and Dt represents the normalized deviation distance in the deviation feature.

8. The method according to claim 1, characterized in that, Based on the aforementioned risk evolution representation for the next 24 hours, calculate the overall risk probability of serious complications occurring in critically ill patients within the next 24 hours post-surgery; including: The next 24 hours are divided into multiple forecast sub-intervals; Based on the risk evolution representation for the next 24 hours, the conditional risk rate of serious complications occurring within each prediction sub-interval is calculated; The conditional hazard rates of each prediction sub-interval are aggregated to obtain the overall risk probability P24h for the next 24 hours; the overall risk probability P24h satisfies: P24h = 1 - ∏(1-h(t+τ)); Where h(t+τ) represents the conditional risk rate of serious complications occurring in the τ-th prediction sub-interval in the future.

9. A system for predicting the risk of serious complications in critically ill patients within 24 hours after surgery, characterized in that, include: The data acquisition module is used to acquire preoperative stable period data and postoperative continuous multimodal health data of critically ill patients; The data preprocessing module is used to preprocess the preoperative stable period data and the postoperative continuous multimodal health data to obtain standardized time-series data. The feature construction module extracts current state features, trend evolution features, fluctuation evolution features, cross-index coupling evolution features, and symptom evolution features from the standardized time series data based on multiple consecutive historical time windows prior to the current moment, and constructs the state vector of the current postoperative state of critically ill patients and the historical state sequence. The baseline modeling module constructs an individualized health baseline model based on the preoperative stable period data. The deviation analysis module is used to compare the state vector with the individualized health baseline model to obtain the deviation characteristics of the current postoperative state of the critically ill patient relative to the preoperative stable state. The risk evolution modeling module constructs a risk evolution representation for the next 24 hours based on the state vector, the historical state sequence, the deviation feature, the trend evolution feature, the fluctuation evolution feature, the cross-indicator coupling evolution feature, and the symptom evolution feature. The 24-hour risk prediction module is used to calculate the overall risk probability of serious complications occurring in critically ill patients within the next 24 hours after surgery, based on the 24-hour risk evolution representation.

10. The system according to claim 9, characterized in that, The postoperative continuous multimodal health data includes vital sign data and subjective symptom data.

11. The system according to claim 9, characterized in that, The data preprocessing module is used to preprocess the preoperative stable period data and the postoperative continuous multimodal health data. The preprocessing includes time alignment, outlier suppression, noise filtering, missing value completion, and / or signal quality assessment.

12. The system according to claim 9, characterized in that, The feature construction module extracts current state features, trend evolution features, fluctuation evolution features, cross-indicator coupling evolution features, and symptom evolution features from the standardized time series data based on multiple consecutive historical time windows prior to the current moment. The current state characteristics include statistical characteristics, trend characteristics, fluctuation characteristics, circadian rhythm characteristics, scene correction characteristics, and subjective symptom characteristics.

13. The system according to claim 12, characterized in that, Obtaining the cross-index coupling evolution characteristics includes: The coupling matrix construction module constructs a cross-index coupling matrix based on the standardized multimodal observation matrix within the current time window; The coupling deviation module is used to obtain coupling deviation characteristics by comparing the difference between the cross-index coupling matrix in the current time window and the preoperative stable period reference coupling matrix; The cross-index coupling evolution feature acquisition module obtains the cross-index coupling evolution feature based on the changing trend of the coupling deviation feature in multiple consecutive historical time windows.

14. The system according to claim 9, characterized in that, The baseline modeling module includes: The individualized health baseline model is constructed based on monitoring data of the patient during a continuous predetermined period of preoperative stabilization. It is divided into scenarios according to resting state, activity state and sleep state, and multiple sub-baseline models are constructed according to daytime and nighttime periods. Each sub-baseline model includes a baseline mean vector, a baseline covariance matrix and a low-dimensional embedding center vector.

15. The system according to claim 14, characterized in that, The deviation analysis module includes: Select the matching sub-baseline model based on the current physiological state and the current time period; The normalized deviation of the current postoperative condition from the preoperative stable baseline was calculated as follows: Dt=sqrt((St-μ(g,p))T·Σ(g,p)^(-1)·(St-μ(g,p))); Where St represents the current postoperative state vector, μ(g,p) represents the baseline mean vector of the corresponding state-time sub-baseline model, Σ(g,p) represents the baseline covariance matrix of the corresponding state-time sub-baseline model, and Dt represents the normalized deviation distance in the deviation feature.

16. The system according to claim 9, characterized in that, The risk prediction module for the next 24 hours includes: The next 24 hours are divided into multiple forecast sub-intervals; Based on the risk evolution representation for the next 24 hours, the conditional risk rate of serious complications occurring within each prediction sub-interval is calculated; The conditional hazard rates of each prediction sub-interval are aggregated to obtain the overall risk probability P24h for the next 24 hours; the overall risk probability P24h satisfies: P24h = 1 - ∏(1-h(t+τ)); Where h(t+τ) represents the conditional risk rate of serious complications occurring in the τ-th prediction sub-interval in the future.

17. A computer device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the method according to any one of claims 1 to 8.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 8.

Images