Early warning time window identification method based on portal vein blood time series data

Abstract

The present application relates to an early warning time window identification method based on portal vein blood time series omics data, comprising: obtaining portal vein blood omics data of a subject at multiple time points, corresponding sampling state data and clinical state data, and forming portal vein blood time series omics data according to the sampling time; determining whether there is abnormal change at each time point according to a preset abnormality determination rule containing individual baseline, reference interval or risk score threshold; when the abnormal change does not satisfy the sampling disturbance determination condition, the corresponding time point is taken as a candidate starting point, and when there is consistent abnormal change in the early warning direction at the subsequent adjacent time point and the clinical termination condition is not satisfied, the starting point of the early warning time window is determined; after the starting point, if the abnormal change continuously does not satisfy the preset abnormality determination rule, or the clinical state data indicates that the clinical diagnosis is established or the clinical endpoint event occurs, the endpoint of the early warning time window is determined, and the early warning time window is output according to the starting point and the endpoint.

Description

Technical Field

[0001] This invention relates to the field of biomedical data processing technology, specifically to an early warning time window identification method based on portal vein blood time series omics data. Background Technology

[0002] Existing technologies have proposed several technical solutions for portal vein-related detection, blood collection, and medical data analysis. For example, publication number CN121667851A discloses a navigation system, medium, and method for portal vein and its tributary vein puncture, which mainly solves the problems of unintuitive positioning of portal vein and its tributary vessels, difficulty in identifying deep vessels, and insufficient criteria for puncture success through image recognition, path planning, or navigation prompts. Another example is publication number CN117693314A, which discloses a blood collection system and method, focusing on blood sample extraction through blood collection execution structure and pressure control. The above solutions can improve the operability of vessel positioning, puncture, or blood collection to a certain extent, but their focus is still mainly on the establishment of sampling pathways, control of sampling actions, or optimization of sampling equipment structure. They have not further solved the problems of how to continuously interpret portal vein blood mics data at multiple time points after sampling, or how to identify early warning stages of diseases from dynamic abnormal changes.

[0003] In clinical research and disease surveillance, portal venous blood can reflect changes in gastrointestinal, pancreatic, spleen, and liver metabolism, inflammation, immunity, or molecular levels at an early stage. Omics testing can also provide multi-dimensional indicators from the metabolomics, proteomics, transcriptomics, or other molecular levels. However, current technologies often use omics testing results as a diagnostic aid at a single time point, or only use whether a certain indicator exceeds the reference range as an abnormal indication, lacking comprehensive processing of the temporal relationship between multiple sampling time points of the same subject, the continuity of abnormal direction, and changes in clinical status. Especially when early pathological changes have not yet reached the clinical diagnostic criteria, omics abnormalities may present as transient fluctuations, cross-omics sequential occurrences, progressively increasing abnormalities in the same direction, or false abnormalities caused by sampling factors. If only a single-point threshold is relied upon for judgment, it is easy to have premature or delayed warnings, or to mistakenly identify sampling disturbances as real pathological changes.

[0004] Furthermore, the portal vein blood sampling process itself may be affected by factors such as hemolysis, abnormal anticoagulation, infusion effects, and uneven sampling intervals, causing the test results at a certain time point to deviate from the individual's baseline or reference range. While existing publicly available technologies can improve the blood collection or puncture process, they do not establish early warning time window identification rules for the correlation between sampling status data, clinical status data, and omics time-series data, nor do they clearly define how to determine the start and end points of the warning time window when an abnormality begins, continues, occurs at a clinical endpoint, or disappears. Therefore, it is necessary to provide an early warning time window identification method that can determine the persistence, directionality, and termination conditions of abnormal changes based on multi-time-point omics data of portal vein blood, combined with sampling status and clinical status, to improve the reliability, interpretability, and clinical applicability of early warning results. Summary of the Invention

[0005] The purpose of this invention is to provide an early warning time window identification method based on portal vein blood time series omics data, thereby solving some of the drawbacks and shortcomings pointed out in the background art.

[0006] The present invention adopts the following technical solution to solve the above-mentioned technical problems: The portal vein blood omics data, corresponding sampling status data and clinical status data of subjects were obtained at multiple time points, and portal vein blood time series omics data were formed according to the sampling time. Based on preset anomaly judgment rules including individual baseline, reference interval or risk score threshold, it is determined whether there are abnormal changes in portal vein blood mics data at each time point; when there is an abnormal change at a time point, and the abnormal change does not meet the sampling perturbation judgment condition, the time point is used as a candidate starting point; if there are abnormal changes with the same warning direction in a preset number of adjacent time points after the candidate starting point, and the clinical status data indicates that the preset clinical termination condition is not met, then the candidate starting point is determined as the starting point of the early warning time window. After the start point of the early warning time window, if the abnormal changes no longer meet the preset abnormality judgment rules at a preset number of adjacent time points, or if the clinical status data indicates that the clinical diagnosis is established or the clinical endpoint event has occurred, then the corresponding time point is determined as the end point of the early warning time window, and the early warning time window is output according to the start point and the end point of the early warning time window.

[0007] Furthermore, when determining the sampling disturbance, if the abnormal change occurs at a single time point and does not continue to adjacent time points, and the time point where the abnormal change occurs meets the preset time correlation conditions with the hemolysis, anticoagulation abnormality, infusion or sampling interval abnormality in the sampling status data, then the abnormal change is determined to be a sampling disturbance, and the corresponding time point is prohibited from being used as a candidate starting point.

[0008] Furthermore, the abnormal changes with the same warning direction are called cross-omics relay anomalies. Cross-omics relay anomalies refer to the occurrence of a second omics category abnormal change with the same warning direction as the abnormal change of the first omics category in a preset number of adjacent time points after the candidate starting point, after the abnormal change of the first omics category appears. The cross-omics relay anomaly is used as the basis for determining the starting point of the early warning time window.

[0009] Furthermore, when determining the end point of the early warning time window, if the abnormal change no longer meets the preset abnormality judgment rule at a preset number of adjacent time points, and then the same abnormal change reappears within a preset observation interval, the time point at which the abnormal change no longer meets the preset abnormality judgment rule will not be determined as the end point of the early warning time window. Instead, the abnormal time periods before and after the disappearance will be merged into the same early warning time window.

[0010] Furthermore, when determining the cross-omics relay anomaly, if the abnormal change in the second omics category occurs after the abnormal change in the first omics category, and there is no intermediate time point between them that fails to meet the preset anomaly determination rule and recovers to the preset baseline interval or reference interval, then the cross-omics relay anomaly is considered to have a continuous relay attribute.

[0011] Furthermore, abnormal changes in the first omics category are marked as leading anomalies, and abnormal changes in the second omics category are marked as response anomalies; when the response anomaly occurs, the candidate starting point where the leading anomaly is located is determined as the starting point of the early warning time window.

[0012] Furthermore, if an abnormal change with the opposite direction to the abnormal change warning of the first omics category occurs in a preset number of adjacent time points after the candidate starting point, the candidate starting point will not be determined as the starting point of the early warning time window for the time being; if a second omics category abnormal change with the same warning direction as the abnormal change of the first omics category occurs again within a preset confirmation interval, the candidate starting point will be determined as the starting point of the early warning time window; if the second omics category abnormal change does not occur again within the preset confirmation interval, the candidate starting point will be cancelled.

[0013] Furthermore, the preset observation interval is determined based on the duration of the last unidirectional abnormal change before the abnormal change subsides; there is a preset mapping relationship between the duration and the preset observation interval, with different duration intervals corresponding to different observation intervals, so as to include the unidirectional abnormal change that reappears within the corresponding observation interval after the abnormal change subsides into the same early warning time window.

[0014] Furthermore, after merging the abnormal time periods before and after the anomaly subsides, a continuity marker is set for the same early warning time window; when the continuity marker exists, the early warning cancellation result is not output at the time point of the anomaly's subsidence.

[0015] Furthermore, when the recurring abnormal change in the same direction appears within the preset observation interval after the abnormal change has subsided, the time point at which the recurring abnormal change occurs is taken as the new observation starting point, and the observation interval is re-determined according to the preset mapping relationship; until the abnormal change does not recur within the re-determined observation interval, the corresponding time point is determined as the end point of the early warning time window.

[0016] This invention acquires portal vein blood omics data, corresponding sampling status data, and clinical status data from multiple time points in subjects, and forms portal vein blood time-series omics data according to sampling time. This allows early warning judgments to no longer rely on single test results, but rather on continuous time series analysis. By introducing preset abnormality judgment rules that include individual baselines, reference intervals, or risk score thresholds, it can adapt to different subjects, different omics indicators, and different clinical risk scenarios, improving the targeting and accuracy of abnormal change identification.

[0017] This invention, when determining the starting point of the early warning time window, does not rely solely on a single point in time for anomaly as the basis for warning. Instead, it uses a point in time as a candidate starting point if the abnormal change does not meet the sampling disturbance judgment criteria. Furthermore, it requires that there be abnormal changes with the same warning direction among a predetermined number of adjacent time points after the candidate starting point, while clinical data indicates that the predetermined clinical termination conditions have not been met. This effectively eliminates occasional fluctuations caused by non-disease factors such as hemolysis, abnormal anticoagulation, abnormal infusion or sampling intervals, reduces the risk of false positive warnings, and facilitates the identification of early abnormal stages with continuous significance before a clinical diagnosis is established.

[0018] This invention further determines the end point of the early warning time window after the initial start point, based on whether the abnormal changes continuously no longer meet the preset abnormality judgment rules, or whether clinical status data indicates the establishment of a clinical diagnosis or the occurrence of a clinical endpoint event. This results in an early warning time window output with clearly defined start and end boundaries. This approach provides clear time-dimensional data for subsequent clinical observation, intervention timing selection, risk stratification, and efficacy evaluation, improving the utilization efficiency and interpretation consistency of portal vein blood time-series data in early disease warning. Attached Figure Description

[0019] Figure 1 This is a flowchart of the early warning time window identification method of the present invention.

[0020] Figure 2 This is a graph showing the changes in portal vein blood time-series omics data in Embodiment 1 of the present invention.

[0021] Figure 3 This is a candidate starting point anomaly determination diagram in Embodiment 1 of the present invention.

[0022] Figure 4 This is a diagram confirming the cross-omics relay in Embodiment 1 of the present invention.

[0023] Figure 5 This is a confirmation diagram of risk score fading and reappearance in Embodiment 2 of the present invention.

[0024] Figure 6 This is a mapping diagram of duration and observation interval in Embodiment 2 of the present invention.

[0025] Figure 7 This is a diagram showing the merging of warning time windows and determination of the endpoint in Embodiment 2 of the present invention. Detailed Implementation

[0026] As attached Figure 1 As shown, in one specific implementation, portal vein blood omics data of the subjects at multiple sampling time points are first acquired, and sampling status data and clinical status data corresponding to each sampling time point are acquired simultaneously. Portal vein blood omics data may include metabolomics data, proteomics data, transcriptomics data, or other omics data that can reflect the biological status of portal vein blood. Sampling status data is used to record whether there are factors that may affect the test results during the sampling process, and clinical status data is used to record the diagnostic status, clinical observation status, and clinical endpoint event status of the subjects at the corresponding time points. The system establishes a correspondence between sample number, sampling time, omics category, test batch, sampling status data, and clinical status data for each sampling time point. After removing samples whose sampling time cannot be confirmed or cannot be matched with the subjects, the portal vein blood omics data are arranged according to the chronological order of the sampling time points to form portal vein blood time-series omics data.

[0027] When generating portal vein blood time-series omics data, the system can also perform quality control processing on data within the same omics category. Quality control processing includes identifying missing data, low-quality test data, and batch-labeled anomalous data. For missing data, if the missing data does not affect the time-series determination of adjacent time points, the system retains the sampling and clinical status records for that time point and marks the corresponding omics item as unusable for anomaly determination; if all key omics items are missing at the same time point, the system marks that time point as an invalid omics time point. For low-quality test data or batch-labeled anomalous data, the system combines test quality labels and sampling status data for screening to avoid mistaking test quality issues for genuine omics anomalies.

[0028] After generating portal vein blood time-series omics data, the system performs anomaly detection on the portal vein blood omics data at each time point according to preset anomaly detection rules. These preset anomaly detection rules may include rules based on changes in the individual baseline of the subjects, rules based on deviations from reference intervals, or rules based on increased risk based on risk score thresholds. The individual baseline of the subjects can be determined by the initial sampling time point data where the subjects have not experienced a clinical endpoint event and their sampling status is normal, or by the subjects' historical stable period data; the reference interval can be determined by healthy control data from the same testing platform, previous non-event sample data, or pre-established clinical reference data; the risk score threshold can be determined by the pre-determined risk stratification results in the training or validation samples. When performing anomaly detection, the system first determines the omics category, anomaly direction, and applicable anomaly detection rules corresponding to the current time point, and then judges whether the portal vein blood omics data at that time point meets the corresponding rules. When the portal vein blood omics data at a certain time point meets the preset anomaly detection rules, the system determines that there is an abnormal change at that time point. Furthermore, the system combines the sampled status data at the corresponding time point to investigate whether the abnormal change was caused by a sampling disturbance. If the abnormal change does not meet the sampling disturbance determination criteria, it indicates that the abnormal change has the potential to serve as an early warning signal, and the system uses the time point where the abnormal change occurs as a candidate starting point.

[0029] After identifying the candidate starting point, the system continues to analyze data from adjacent time points. If there are abnormal changes with a consistent warning direction among a preset number of adjacent time points after the candidate starting point, and the clinical status data indicates that the preset clinical endpoint conditions have not yet been met, the system determines the candidate starting point as the starting point of the early warning time window. The preset number of adjacent time points can be pre-set as one or more adjacent time points based on the sampling frequency; when the sampling frequency is high, two or more consecutive adjacent time points can be used as the confirmation criterion, and when the sampling frequency is low, the next adjacent time point or the next effective omics time point can be used as the confirmation criterion. Consistent warning direction means that abnormal changes at multiple time points all point to the same risk change direction, such as all pointing to an increased risk direction. This avoids misjudging isolated, sporadic abnormal changes as early warning starting points.

[0030] After determining the starting point of the early warning time window, the system continues to monitor portal vein blood omics data and clinical status data at subsequent time points. If the abnormal change no longer meets the preset abnormality judgment rules at a preset number of adjacent time points, the system determines that the abnormal change has subsided and designates the corresponding time point as the end point of the early warning time window. If the clinical status data indicates that a clinical diagnosis has been established or a clinical endpoint event has occurred, the system also designates the corresponding time point as the end point of the early warning time window. Finally, based on the starting and ending points of the early warning time window, the system generates and outputs the early warning time window to characterize the warning time range from the onset of omics abnormality to its subsidence or the occurrence of a clinical event in the subjects.

[0031] Based on the above specific implementation method, when determining sampling disturbances, the system jointly analyzes the temporal continuity of abnormal changes with sampling status data. If an abnormal change only occurs at a single time point and does not continue to occur at adjacent sampling time points, the system marks the abnormal change as an anomaly to be investigated. Subsequently, the system reads the sampling status data corresponding to the time point where the abnormal change occurs and determines whether there are sampling-related factors such as hemolysis, anticoagulation abnormalities, infusion effects, or abnormal sampling intervals at that time point. Sampling status data may include one or more of the following: sample collection time, sample processing time, anticoagulant type, sample appearance record, whether hemolysis occurred, whether infusion was received, whether there was a sampling delay, whether there was duplicate sampling, and batch marking.

[0032] When the time point of the anomaly to be investigated meets the preset time correlation conditions with the aforementioned sampling-related factors, the system determines that the abnormal change is more likely caused by the sampling process than by a real change in the subject's portal vein blood proteomics status. The preset time correlation conditions may include the occurrence time of the sampling-related factors being the same as the time point of the abnormal change, or the occurrence time of the sampling-related factors being within a preset adjacent sampling range. For abnormal sampling intervals, the system can mark time points that significantly deviate from the preset sampling plan and cannot form a continuous temporal relationship with preceding and following sampling time points as abnormal sampling interval time points; for infusion effects, the system can mark cases where the infusion record is adjacent to the sampling time and the abnormal change at that time point does not continue at subsequent time points as infusion-related disturbances; for hemolysis or anticoagulation abnormalities, the system can mark cases where corresponding records exist in the sampling status data and the abnormal change is limited to that time point as sample quality-related disturbances. Based on this, the system determines the abnormal change as a sampling disturbance and excludes that time point during the early warning time window identification process, preventing it from being identified as a candidate starting point, thereby reducing the interference of single-point sampling anomalies on early warning results.

[0033] Based on the above specific implementation, after determining the candidate starting point, the system performs cross-omics continuity analysis on adjacent time points after the candidate starting point. Specifically, when an abnormal change in the first omics category first appears at the candidate starting point, the system uses this abnormal change as an initial warning signal and continues to read data from other omics categories in a preset number of adjacent time points after the candidate starting point. If an abnormal change in the second omics category appears in the above adjacent time points, and the abnormal change in the second omics category has the same warning direction as the abnormal change in the first omics category, then the system identifies this time sequence relationship as a cross-omics relay anomaly. The same warning direction can indicate that both the abnormal change in the first omics category and the abnormal change in the second omics category point to an increased risk direction, or both point to the same preset risk state. The system can pre-set the correspondence between abnormal items and warning directions for each omics category. For example, abnormal changes related to increased inflammation, aggravated metabolic disorders, enhanced tissue damage, or abnormal immune activation can be configured as an increased risk direction, and abnormal changes related to the weakening or recovery of the above states can be configured as a decreased risk direction. The system uses cross-omics relay anomalies as an important basis for determining whether a candidate starting point can be identified as the starting point of the early warning time window, thereby avoiding relying solely on isolated anomalies of a single omics category for early warning judgment.

[0034] When identifying cross-omics relay anomalies, the system further determines whether there is a continuous relay attribute between the anomalous changes in the first and second omics categories. If the anomalous change in the second omics category occurs after the anomalous change in the first omics category, and there is no intermediate time point between them that no longer meets the preset anomaly judgment rules and has recovered to the preset baseline or reference interval, it indicates that no effective recovery has occurred between the anomalous changes in the first and second omics categories. In this case, the system determines that the cross-omics relay anomaly has a continuous relay attribute. If there is an invalid omics time point between the anomalous changes in the first and second omics categories, the system does not directly identify the invalid omics time point as an anomalous recovery, but continues to read the valid omics time points before and after it, and determines whether the continuous relay attribute is maintained based on whether the warning directions of the valid omics time points before and after are consistent. In this way, a truly continuously developing omics anomaly process can be distinguished from an independent anomaly process that has recovered and then recurred.

[0035] During the identification process, the system marks abnormal changes in the first omics category as lead anomalies and abnormal changes in the second omics category as response anomalies. Lead anomalies represent the earliest occurring anomaly signal that triggers the candidate starting point determination, while response anomalies represent anomalies that form a unidirectional response to the lead anomaly at subsequent adjacent time points. When the system detects a response anomaly, it indicates that the lead anomaly at the candidate starting point has been confirmed by unidirectional abnormal changes in subsequent omics categories. At this point, the system determines the candidate starting point where the lead anomaly is located as the starting point of the early warning time window. The system can also record the lead omics category, response omics category, lead anomaly item, response anomaly item, and warning direction corresponding to the starting point of this early warning time window for subsequent output of the early warning time window results.

[0036] If the system detects an abnormal change in the opposite direction to the warning direction of the first omics category within a preset number of adjacent time points after the candidate starting point, it indicates that the warning direction corresponding to the candidate starting point has not yet been stably confirmed. In this case, the system temporarily does not designate the candidate starting point as the starting point of the early warning time window and proceeds to the continued observation process within a preset confirmation interval. The preset confirmation interval can be set to one or more subsequent effective omics time points according to the sampling frequency, or it can be set to a predetermined observation duration according to clinical observation needs. Within the preset confirmation interval, if a second omics category abnormal change with the same warning direction as the first omics category abnormal change reappears, the system determines that the leading abnormality corresponding to the candidate starting point has been confirmed and designates the candidate starting point as the starting point of the early warning time window. If no second omics category abnormal change reappears within the preset confirmation interval, the system cancels the candidate starting point to reduce the risk of misjudgment caused by abnormal changes in the opposite direction to the early warning time window identification results.

[0037] Based on the above specific implementation method, when determining the end point of the early warning time window, the system does not directly use the time point when the abnormal change first subsides as the end point, but rather confirms it in conjunction with subsequent observation results. Specifically, when the abnormal change no longer meets the preset abnormality judgment rules at a preset number of adjacent time points, the system first identifies this state as an abnormal change subsidence state and initiates a preset observation interval. Within the preset observation interval, the system continues to monitor portal vein blood time series data. If an abnormal change with the same warning direction as the abnormal change before subsidence reappears within this observation interval, it indicates that the abnormal change is a continuation process after a short-term fluctuation, rather than the end of the early warning state. At this time, the system does not determine the time point when the abnormal change no longer meets the preset abnormality judgment rules as the end point of the early warning time window, but instead merges the abnormal time period before subsidence and the abnormal time period after reappearance into the same early warning time window.

[0038] The preset observation interval is determined based on the duration of the last homologous abnormal change before the abnormal change subsides. The system pre-sets a mapping relationship between duration and observation interval, so that different duration intervals correspond to different observation intervals. For example, when the abnormal change lasts for only one sampling interval, the system can use the next effective omics time point as the observation interval; when the abnormal change lasts for two or more sampling intervals, the system can use the next two or more effective omics time points as the observation interval; when clinical status data indicates that the subject is in a high-risk observation state, the system can extend the observation interval. The above observation intervals are pre-configured before the system runs and remain consistent during the identification process. When the duration before the abnormal change subsides is within the corresponding duration interval, the system calls the observation interval corresponding to that duration interval and uses this observation interval as the basis for determining whether the abnormal change continues to belong to the same early warning time window. In this way, homologous abnormal changes that reappear within the corresponding observation interval after subsiding can be included in the same early warning time window, thereby avoiding premature termination of the warning time window due to short-term weakening or temporary disappearance of the abnormal change.

[0039] After merging the abnormal time periods before and after their disappearance, the system sets a continuity marker for the merged early warning time window. The continuity marker indicates that an abnormal change reappears after a short period of disappearance within the early warning time window. When the continuity marker is present, the system does not output an early warning cancellation result at the time point of the abnormal change's disappearance, but continues to observe subsequent time points. When a reappearance of the same-direction abnormal change after disappearance occurs within a preset observation interval, the system uses the time point of the reappearance of the same-direction abnormal change as a new observation starting point and redetermines the observation interval according to a preset mapping relationship. Subsequently, the system continues to determine whether the same-direction abnormal change reappears within the newly determined observation interval. Only when the abnormal change does not reappear within the newly determined observation interval does the system determine the corresponding time point as the end point of the early warning time window.

[0040] When outputting the early warning time window, the system can output the subject identifier, the start point of the early warning time window, the end point of the early warning time window, the omics category of the triggering start point, the response omics category of the confirmation start point, the warning direction, whether there are sampling perturbation exclusion records, and whether there are continuity markers. If the end point of the early warning time window is triggered by the establishment of a clinical diagnosis or the occurrence of a clinical endpoint event, the system will mark the endpoint type as clinical event triggered; if the end point of the early warning time window is triggered by the fading of abnormal changes and confirmation by the observation interval, the system will mark the endpoint type as abnormal fading triggered. By outputting the above information, the source of the start point, the source of the end point, and the intermediate confirmation process of the early warning time window can be clearly recorded, facilitating subsequent review and application.

[0041] Example 1: In one specific implementation, the system acquires portal vein blood mics data, sampling status data, and clinical status data for person A at four sampling time points: T0, T1, T2, and T3, and generates portal vein blood time-series mics data according to the sampling time. The portal vein blood mics data includes metabolomics data, proteomics data, and transcriptomics data; the sampling status data includes hemolysis status, anticoagulation status, infusion records, and sampling interval records; and the clinical status data includes records of whether preset clinical termination conditions have been met at the corresponding time points.

[0042] In this embodiment, T0 is the individual's baseline time point, and T1, T2, and T3 are subsequent adjacent sampling time points. The system recorded the following data at T0: portal vein blood lactate concentration was 1.2 mmol / L, CRP level was 5.8 mg / L, and inflammatory pathway expression score was 0.32; at T1, lactate concentration was 2.1 mmol / L, CRP level was 6.1 mg / L, and inflammatory pathway expression score was 0.35; at T2, lactate concentration was 2.4 mmol / L, CRP level was 18.6 mg / L, and inflammatory pathway expression score was 0.51; at T3, lactate concentration was 2.5 mmol / L, CRP level was 24.2 mg / L, and inflammatory pathway expression score was 0.63. Clinical status data at each time point showed that the preset clinical termination criteria had not yet been met. Figure 2 The expression scores of lactate, CRP and inflammatory pathways were normalized and displayed according to their respective judgment criteria, so that the data changes of different omics categories could be compared on the same time axis; among them, the shaded area in the T1 to T2 interval represents the process in which metabolomics lead abnormality and proteomics response abnormality form a relay in the same direction at adjacent time points.

[0043] The system performs anomaly detection on portal vein blood proteomics data at each time point according to preset anomaly detection rules. For metabolomics lactate concentration, the system uses the percentage change relative to the individual baseline as the anomaly detection criterion, calculated as follows:

[0044] in, Indicates the first The proportion of change at each sampling time point relative to the individual baseline Indicates the first The omics index detection values ​​at each sampling time point This represents the individual baseline value at time point T0. This formula is used to convert ratios into percentages. The derivation process involves first calculating the difference between the current time point's measured value and the individual's baseline value, then dividing this difference by the individual's baseline value to obtain the increase in the current measured value relative to the individual's baseline value. The system is pre-set to determine an abnormal change in the direction of increased risk for this metabolomics indicator when the lactate concentration increases by at least 50% relative to the individual's baseline.

[0045] Substituting the lactate concentration at time T1 into the above calculation formula, we get:

[0046] Therefore, it can be seen that the lactate concentration at time point T1 increased by 75% relative to the baseline of individual T0, reaching and exceeding the preset 50% abnormality judgment threshold. The system judged that there was a metabolomics abnormal change in the direction of increased risk at time point T1. Figure 3 The changes in lactate relative to the individual baseline at time points T0 to T3 are further shown. At time point T1, the 50% anomaly threshold was first exceeded and the area was marked as a candidate starting point by the shaded region. Time points T2 and T3 correspond to changes of 100% and 108%, respectively, to indicate that the abnormal change continued in subsequent time points.

[0047] The system further reads the sampling status data at time point T1 and confirms that no hemolysis, anticoagulation abnormalities, infusion effects, or sampling interval abnormalities are recorded at this time point. Therefore, this abnormal change does not meet the sampling disturbance judgment criteria, and the system uses time point T1 as a candidate starting point. When judging sampling disturbances, if an abnormal change occurs only at a single time point and does not continue to adjacent time points, and the time point where the abnormal change occurs meets the preset time correlation conditions with hemolysis, anticoagulation abnormalities, infusion effects, or sampling interval abnormalities in the sampling status data, then the system judges the abnormal change as a sampling disturbance and prohibits this time point from being a candidate starting point. In this embodiment, there is still an abnormal change in the direction of increased risk at time point T2 after T1, and the sampling status data at T1 is normal; therefore, the abnormal change at time point T1 is not excluded.

[0048] After identifying T1 as a candidate starting point, the system continued to read data from other omics categories from a predetermined number of adjacent time points after the candidate starting point. At time point T2, the proteomics CRP level was 18.6 mg / L, exceeding the upper limit of the predetermined reference range of 10.0 mg / L. The system determined that there was an abnormal proteomics change at time point T2 indicating an increased risk. Since the abnormal change at time point T2 occurred after the abnormal metabolomics change at time point T1, and both pointed towards an increased risk, the system marked the abnormal metabolomics change at time point T1 as a leading abnormality and the abnormal proteomics change at time point T2 as a response abnormality.

[0049] To further confirm the anomaly in the cross-omics relay, the system calculates the cross-omics relay confirmation score, which is calculated as follows:

[0050] in, This indicates the confirmed score for the cross-omics relay. The value indicates the consistency of the warning direction: 1 is taken when the two consecutive abnormal changes in omics both point to the same risk direction, and 0 is taken otherwise. This indicates the value of the continuous relay attribute. It is set to 1 if there is no intermediate time point between the leading anomaly and the response anomaly that recovers to the preset baseline or reference interval; otherwise, it is set to 0. and These represent the weights of directional consistency and continuous relay attributes, respectively. In this embodiment, the system pre-sets... , and set The cross-omics relay anomaly is confirmed in time. The derivation of this formula is to break down the cross-omics relay judgment into two components: whether the warning direction is consistent and whether the time sequence is continuous, and then obtain the confirmation score by summing the weights.

[0051] In this embodiment, both the abnormal lactate level at time T1 and the abnormal CRP level at time T2 point in the direction of increased risk, therefore There is no intermediate time point between T1 and T2 where the individual has recovered to the baseline or reference range. Substituting the above values ​​into the calculation formula, we get:

[0052] Since the cross-omics relay confirmation score reached the preset confirmation conditions, the system determined that the leading anomaly at time point T1 and the response anomaly at time point T2 constituted a cross-omics relay anomaly with continuous relay attributes. Figure 4 The directional consistency, continuous relay attribute, and confirmation score are represented by judgment values, where directional consistency is valued at 1.0, continuous relay attribute at 1.0, and confirmation score reaching the confirmation threshold of 1.0. This figure illustrates that cross-omics relay is not determined solely by an increase in a single indicator, but is jointly confirmed by two judgment conditions: a change in directional risk and the absence of effective recovery. At this point, the system determines the T1 time point, where the lead anomaly occurred, as the starting point of the early warning time window.

[0053] If the system detects an abnormal change in the opposite direction to the warning direction of the first omics category within a preset number of adjacent time points after the candidate starting point, the candidate starting point will not be temporarily designated as the starting point of the early warning time window, and the system will continue to observe within a preset confirmation interval. Within the preset confirmation interval, if a second omics category abnormal change with the same warning direction as the first omics category abnormal change occurs again, the system determines that the precursor abnormality corresponding to the candidate starting point has received a response abnormality confirmation and designates the candidate starting point as the starting point of the early warning time window; if the aforementioned second omics category abnormal change does not occur again, the system cancels the candidate starting point. In this embodiment, no abnormality with the opposite warning direction occurs at time point T2, therefore, there is no need to proceed with the candidate starting point cancellation process.

[0054] The system further analyzed the transcriptomic expression score of inflammatory pathways at time point T3, which increased from 0.32 at T0 to 0.63 at T3, consistent with the increased risk trends observed at T1 and T2. Combined with... Figure 2 The trend of the inflammation score continuously increasing relative to the threshold at time point T3 further indicates that there are abnormal changes in the same direction at multiple adjacent time points after T1, and the clinical status data shows that the preset clinical termination conditions have not yet been met. Therefore, the determination of T1 as the starting point of the early warning time window is maintained.

[0055] Through the above processing, in the portal vein blood time-series omics data of person A, the system first identifies metabolomics precursor abnormalities at time point T1 based on the individual baseline, then excludes sampling perturbation factors, and confirms cross-omics relay abnormalities by combining the abnormal proteomics response at time point T2, and finally determines time point T1 as the starting point of the early warning time window. Figure 2 Used to demonstrate the co-directional changes in time series data from multiple omics systems. Figure 3 Used to demonstrate the anomaly detection process for candidate starting points. Figure 4 This method is used to demonstrate the confirmation process of cross-omics relay anomalies. It enables the identification of early warning time windows by utilizing continuous, unidirectional abnormal changes in portal vein blood time-series omics data before clinical termination criteria are met.

[0056] Example 2: In another specific implementation, after determining the starting point of the early warning time window for individual A, the system continues to acquire portal vein blood profilometry data, clinical status data, and abnormal status records of the subject at subsequent time points, and determines the end point of the early warning time window based on the persistence, regression, and recurrence of subsequent abnormal changes. Portal vein blood profilometry data can be converted into a normalized risk score between 0 and 1; a higher risk score indicates that the portal vein blood profilometry status at the corresponding time point is closer to the preset elevated risk state.

[0057] In this embodiment, the system uses postoperative day 0 as the starting point of the early warning time window, and postoperative days 1 to 6 as subsequent monitoring time points. The system pre-sets a risk score threshold of 0.65. When the risk score at a certain time point is not lower than 0.65, it is determined that there is an abnormal change in the direction of risk increase at that time point; when the risk score is lower than 0.65, it is determined that the time point does not meet the preset abnormal judgment rules. Clinical status data is used to record whether a clinical diagnosis has been established or whether a clinical endpoint event has occurred at the corresponding time point.

[0058] The system recorded the portal vein blood risk scores of patient A from postoperative day 0 to day 6 as 0.72, 0.76, 0.74, 0.58, 0.71, 0.62, and 0.57, respectively. From postoperative day 0 to day 2, the risk scores were all above 0.65, indicating a persistent abnormal change in the direction of increasing risk during this period. On postoperative day 3, the risk score was 0.58, below 0.65, suggesting the abnormal change had initially subsided. On postoperative day 4, the risk score was 0.71, again above 0.65, indicating a recurrence of an abnormal change with the same warning direction as before the abnormal change had subsided. On postoperative days 5 and 6, the risk scores were 0.62 and 0.57, respectively, both below 0.65. Clinical status data at each time point showed no clinical endpoint events, and no clinical diagnosis was recorded from postoperative day 0 to day 6. Figure 5 The above time series process is presented using a risk score line, a risk score threshold line, and different shaded intervals. Day 0 to 2 corresponds to the abnormal period before the disappearance, day 3 corresponds to a brief disappearance, day 4 corresponds to the reappearance of the same abnormality, and days 5 to 6 correspond to the endpoint confirmation observation process. This intuitively reflects that the abnormal change does not immediately terminate when it first falls below the threshold.

[0059] When determining whether an abnormal change meets the preset anomaly judgment rules, the system can use the difference between the risk score and the preset threshold as the judgment basis, and the calculation formula is as follows:

[0060] in, Indicates the first The daily threshold difference, Indicates the first Normalized risk score of portal vein blood on a given day This represents the preset risk scoring threshold. The derivation process involves directly comparing the risk score at the current time point with the preset threshold. If the difference is not less than 0, it indicates that the current time point has reached or exceeded the threshold, and the system determines that there is an abnormal change in the direction of risk increase at this time point. If the difference is less than 0, it indicates that the current time point has not reached the threshold, and the system determines that this time point does not meet the preset abnormal judgment rules.

[0061] Substituting the data from the 3rd and 4th days post-surgery into the above calculation formula, we obtain:

[0062]

[0063] The threshold difference on postoperative day 3 was negative. The system identified this time point as a state of regression of abnormal changes, but did not immediately determine it as the end of the early warning time window. The threshold difference on postoperative day 4 was positive, and this abnormal change, like the abnormal changes from postoperative day 0 to day 2, pointed in the direction of increased risk. Therefore, the system determined that postoperative day 4 was a recurrence of the same abnormality after regression.

[0064] After identifying the fading of abnormal changes on the 3rd postoperative day, the system initiates a preset observation interval. The preset observation interval is determined based on the duration of the last abnormal change in the same direction before its fading. Before the fading of abnormal changes, there were abnormal changes in the direction of increased risk from postoperative day 0 to day 2; therefore, the duration before fading was 3 sampling days. The system pre-sets a mapping relationship between duration and observation interval: when the duration is 1 sampling day, the observation interval is 2 subsequent valid sampling days; when the duration is 2 or 3 sampling days, the observation interval is 2 subsequent valid sampling days; when the duration is greater than 3 sampling days, the observation interval is 3 subsequent valid sampling days. Figure 6 The above-mentioned preset relationship is represented by both bar charts and line charts, and the observation interval of calling two subsequent valid sampling days is highlighted when the duration is 3 sampling days and when the duration is 1 sampling day, so that the rule for determining the observation interval can correspond to the subsequent endpoint judgment process.

[0065] In this embodiment, the system uses the following formula to represent the rule for determining the observation interval:

[0066] in, This indicates the preset observation interval, in units of subsequent valid sampling days. This indicates the duration of the last unidirectional abnormal change before the abnormal change subsides, expressed in sampling days. This indicates the pre-configured mapping relationship between the system and the duration. The corresponding observation interval. The derivation process of this formula is to first count the number of sampling days covered by continuous unidirectional abnormal changes before the anomaly subsides, and then call the corresponding observation interval according to the preset mapping relationship so that the duration of the anomaly corresponds to the pre-set endpoint to confirm the observation range.

[0067] Substituting the duration before decay in this embodiment into the above mapping relationship, we get:

[0068] After the abnormality subsided on the 3rd postoperative day, the system continued to observe for two subsequent effective sampling days, namely the 4th and 5th postoperative days. Because an abnormal change in the direction of risk increase reappeared within this observation interval on the 4th postoperative day, the system did not designate the 3rd postoperative day as the end of the early warning time window, nor did it output an early warning cancellation result on the 3rd postoperative day. Instead, it combined the abnormal time period formed from the 0th to the 2nd postoperative day with the abnormal time period that reappeared on the 4th postoperative day into the same early warning time window.

[0069] After merging the abnormal time periods before and after the disappearance, the system sets a continuity marker for this early warning time window. The continuity marker is used to indicate that there are cases where abnormal changes reappear after a short period of disappearance within the early warning time window. When the continuity marker exists, the system continues to observe subsequent time points, rather than using a risk score below a threshold on the 3rd postoperative day as the basis for determining that the early warning time window has ended. Figure 7 The merging logic is represented by an abnormal status bar and an output time window bar. The brief fading on day 3 does not cut off the output time window, and the reappearance of the same direction on day 4 is included in the same time window. The output time window extends from day 0 to the final confirmed day 6.

[0070] When the recurrence of the same-direction abnormal change on the 4th postoperative day occurs within the preset observation interval, the system uses the 4th postoperative day as the new observation starting point and re-determines the observation interval according to the preset mapping relationship. Since the duration of the recurrence of the same-direction abnormal change on the 4th postoperative day is one sampling day, the system determines the new observation interval as two subsequent valid sampling days, namely the 5th and 6th postoperative days, according to the mapping relationship. The calculation results are as follows:

[0071] The risk score was 0.62 on postoperative day 5 and 0.57 on postoperative day 6, both below 0.65. The system determined that no further abnormal changes in the same direction of risk increase occurred within the newly determined observation interval. Substituting the data from postoperative day 5 and day 6 into the threshold difference calculation formula, we obtain:

[0072]

[0073] Since the threshold difference on postoperative days 5 and 6 is negative, and the clinical status data still does not indicate the establishment of a clinical diagnosis or the occurrence of a clinical endpoint event, the system designates postoperative day 6 as the end of the early warning time window. If a similar abnormal change with a risk score of not less than 0.65 occurs again within the redefined observation interval, the system will use the time point of occurrence of the recurrence of the similar abnormal change as the new observation starting point, and continue to redefine the observation interval according to the preset mapping relationship until no similar abnormal change occurs again within the redefined observation interval.

[0074] The system outputs the early warning time window results for person A based on the start and end points of the early warning time window. The start point of the early warning time window is day 0 after surgery, the end point of the early warning time window is day 6 after surgery, the end point type is abnormal regression and confirmed by the observation interval, the continuity is marked as present, and the combined reason is recorded as abnormal regression on day 3 after surgery and reappearance of the same abnormal change on day 4 after surgery. Figure 5 This is used to illustrate the process of risk scores fading and reappearing after crossing a threshold. Figure 6 This is used to illustrate the preset mapping relationship that is invoked based on the observation interval and the duration of the anomaly. Figure 7 This indicates that the abnormal time periods before and after the decline are combined into the same early warning time window, and the endpoint is determined on the 6th day. In this way, the system can avoid prematurely terminating the early warning time window due to a short-term decline in risk score, and can determine the endpoint of the early warning time window after the same-direction abnormal change has not reappeared within the redefined observation interval.

Claims

1. A method for identifying early warning time windows based on portal vein blood time-series omics data, characterized in that, include: The portal vein blood omics data, corresponding sampling status data and clinical status data of subjects were obtained at multiple time points, and portal vein blood time series omics data were formed according to the sampling time. Based on preset anomaly determination rules including individual baseline, reference interval or risk score threshold, it is determined whether there are abnormal changes in portal vein blood mics data at each time point; when there is an abnormal change at a time point, and the abnormal change does not meet the sampling perturbation determination condition, the time point is used as a candidate starting point. If there are abnormal changes in the same direction of warning in a preset number of adjacent time points after the candidate starting point, and the clinical status data indicates that the preset clinical termination conditions are not met, then the candidate starting point will be determined as the starting point of the early warning time window. After the start point of the early warning time window, if the abnormal changes no longer meet the preset abnormality judgment rules at a preset number of adjacent time points, or if the clinical status data indicates that the clinical diagnosis is established or the clinical endpoint event has occurred, then the corresponding time point is determined as the end point of the early warning time window, and the early warning time window is output according to the start point and the end point of the early warning time window.

2. The method for identifying early warning time windows based on portal vein blood time-series omics data according to claim 1, characterized in that, When determining the sampling disturbance, if the abnormal change occurs at a single time point and does not continue to adjacent time points, and the time point where the abnormal change occurs meets the preset time correlation conditions with the hemolysis, anticoagulation abnormality, infusion or sampling interval abnormality in the sampling status data, then the abnormal change is determined to be a sampling disturbance, and the corresponding time point is prohibited from being used as a candidate starting point.

3. The method for identifying early warning time windows based on portal vein blood time-series omics data according to claim 1, characterized in that, The abnormal changes with the same warning direction are called cross-omics relay anomalies. Cross-omics relay anomalies refer to the occurrence of an abnormal change in the first omics category at the candidate starting point, followed by an abnormal change in the second omics category with the same warning direction at a preset number of adjacent time points after the candidate starting point. The cross-omics relay anomalies are used as the basis for determining the starting point of the early warning time window.

4. The method for identifying early warning time windows based on portal vein blood time-series omics data according to claim 1, characterized in that, When determining the end point of the early warning time window, if the abnormal change no longer meets the preset abnormality judgment rule at a preset number of adjacent time points, and then the same abnormal change reappears within a preset observation interval, the time point at which the abnormal change no longer meets the preset abnormality judgment rule will not be determined as the end point of the early warning time window. Instead, the abnormal time periods before and after the disappearance will be merged into the same early warning time window.

5. The method for identifying early warning time windows based on portal vein blood time-series omics data according to claim 3, characterized in that, When determining the cross-omics relay anomaly, if the abnormal change of the second omics category occurs after the abnormal change of the first omics category, and there is no intermediate time point between them that fails to meet the preset anomaly judgment rule and recovers to the preset baseline interval or reference interval, then the cross-omics relay anomaly is considered to have a continuous relay attribute.

6. The method for identifying early warning time windows based on portal vein blood time-series omics data according to claim 3, characterized in that, Abnormal changes in the first omics category are marked as leading anomalies, and abnormal changes in the second omics category are marked as response anomalies. When a response anomaly occurs, the candidate starting point of the leading anomaly is determined as the starting point of the early warning time window.

7. The method for identifying early warning time windows based on portal vein blood time-series omics data according to claim 3, characterized in that, If an abnormal change occurs in a predetermined number of adjacent time points after the candidate starting point, and the abnormal change warning direction is opposite to that of the first omics category, then the candidate starting point will not be determined as the starting point of the early warning time window. If a second omics category abnormal change with the same warning direction as the abnormal change of the first omics category reappears within the preset confirmation interval, the candidate starting point will be determined as the starting point of the early warning time window. If no abnormal change in the second omics category occurs again within the preset confirmation interval, the candidate starting point is cancelled.

8. The method for identifying early warning time windows based on portal vein blood time-series omics data according to claim 4, characterized in that, The preset observation interval is determined based on the duration of the last unidirectional abnormal change before the abnormal change subsides; there is a preset mapping relationship between the duration and the preset observation interval, with different duration intervals corresponding to different observation intervals, so as to include the unidirectional abnormal change that reappears within the corresponding observation interval after the abnormal change subsides into the same early warning time window.

9. The method for identifying early warning time windows based on portal vein blood time-series omics data according to claim 4, characterized in that, After merging the abnormal time periods before and after the disappearance, a continuity marker is set for the same early warning time window; When the continuity marker exists, the early warning cancellation result is not output at the time point when the abnormal change subsides.

10. The method for identifying early warning time windows based on portal vein blood time-series omics data according to claim 8, characterized in that, When a recurring abnormal change in the same direction appears within the preset observation interval after the previous abnormal change has subsided, the time point at which the recurring abnormal change occurs is taken as the new observation starting point, and the observation interval is redefined according to the preset mapping relationship; until the abnormal change does not recur within the redefined observation interval, the corresponding time point is defined as the end point of the early warning time window.

Images