Clinical subject grouping processing method and system based on state detection

Abstract

The present application relates to the technical field of data processing, and particularly relates to a clinical subject grouping processing method and system based on state detection. The method comprises the following steps: acquiring a multi-dimensional state detection data sequence of a target subject within a preset time window and performing time sequence alignment and cleaning; extracting a comprehensive physiological detection value by using a state feature extraction network, and then calculating a time sequence deterioration rate index and an expected steady state offset; based on the expected steady state offset, obtaining an imbalance deterioration degree of the target subject assigned to each candidate test group, and combining a risk assessment index of the target subject to obtain a final assignment probability score, and grouping the target subject based on the final assignment probability score. The present application can capture the implicit disease evolution trend of the subject, effectively intercept improper assignment actions, and guarantee the rationality and balance of the clinical trial grouping process.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for grouping clinical subjects based on state detection. Background Technology

[0002] In the field of clinical trial data processing and subject allocation technology, ensuring the balance of disease distribution among subjects in each experimental group is a core prerequisite for evaluating the effectiveness and safety of the trial. Existing dynamic randomization algorithms and grouping mechanisms generally rely on stratified processing of static, discrete baseline data collected from subjects during the screening period, completely ignoring the differences in physiological evolution of individual subjects over time.

[0003] This traditional data processing model has significant flaws in its logical deduction: it forcibly solidifies the continuously changing physiological indicators of subjects into static values ​​at a single time point, completely losing the characteristics of disease evolution trends and deterioration rates within the critical time window before enrollment, resulting in a severely distorted data foundation upon which clinical decisions rely. When subjects are in a latent deterioration stage, their static baseline indicators may still remain within the apparent safety threshold, causing the system to fail to detect potential risks in a timely manner and to fail to identify high-risk individuals who appear stable but are actually deteriorating rapidly. These flaws can trigger a serious chain of misjudgments, causing patients in the high-risk evolution phase to be incorrectly classified as stable individuals, leading to an overconcentration of high-risk patients in certain experimental groups while other experimental groups are generally in a low-risk state. This seriously violates the principle of balance in randomization, ultimately not only severely disrupting the baseline balance between groups but also easily triggering safety events such as sudden changes in physiological indicators in the later stages of the trial, greatly reducing the credibility and reliability of clinical trial results. Summary of the Invention

[0004] To address the technical problem that existing clinical trial grouping algorithms rely on static discrete data, leading to the loss of disease evolution trends and resulting in the misclassification of high-risk patients and group imbalance, this invention provides a clinical subject grouping method and system based on state detection.

[0005] In a first aspect, the present invention provides a clinical subject grouping method based on state detection, employing the following technical solution: The clinical subject grouping method based on state detection includes the following steps: The state detection data sequence of the target subject in each dimension within a preset time window is obtained. After time alignment, outlier cleaning and normalization, the normalized state detection data sequence of each dimension is obtained. The normalized state detection data sequence of the target subject in each dimension is input into the state feature extraction network to obtain the comprehensive physiological detection value and risk assessment index of the target subject at each sampling time. Based on the comprehensive physiological detection value, the temporal deterioration rate index of the target subject is obtained. Based on the temporal deterioration rate index and the total follow-up period specified in the clinical trial protocol for the target subject's disease, the expected steady-state offset is obtained. Based on the expected steady-state offset, the degree of imbalance worsening in the allocation of target subjects to each candidate experimental group is obtained; based on the degree of imbalance worsening and the risk assessment index of the target subjects, the final allocation probability score of the target subjects to each candidate experimental group is obtained; the target subjects are grouped based on the final allocation probability score of the target subjects to each candidate experimental group.

[0006] The innovation of this invention lies in its ability to effectively capture latent disease progression trends before subject enrollment, avoid assessment omissions and misclassification of high-risk individuals caused by a single static indicator, intercept inappropriate allocation actions that cause imbalance in inter-group distribution, and significantly improve the balance and safety of clinical trial grouping.

[0007] Preferably, the step of acquiring the state detection data sequence of the target subject in each dimension within a preset time window, and then performing time-series alignment, outlier cleaning, and normalization processing to obtain the normalized state detection data sequence for each dimension, includes: Preset time window Every hour, the system connects to the medical information system to obtain the target subject's vital signs data, laboratory test data, and symptom self-report data within the preset time window, forming a data sequence of the target subject's status in various dimensions.

[0008] By connecting with medical information systems to comprehensively collect vital sign data, laboratory test data, and symptom self-report data within a preset time window, a multi-dimensional data matrix covering objective physiological indicators and subjective feelings can be constructed. This avoids the limitations of assessing the true condition of subjects with single-dimensional data and provides rich and comprehensive basic data support for subsequent accurate analysis.

[0009] Preferably, obtaining the comprehensive physiological test values ​​and risk assessment index of the target subject at each sampling time includes: The normalized state detection data sequences of the target subject in each dimension are input into a long short-term memory network for processing to obtain the basic temporal feature matrix of the target subject. The basic temporal feature matrix is ​​then input into the attention mechanism layer to calculate the dynamic weight distribution vector of each dimension at each sampling time. The dynamic weight distribution vector is then weighted and summed with the normalized state detection data of each dimension at each sampling time to output the comprehensive physiological detection value of the target subject at each sampling time. The temporal feature vector composed of the comprehensive physiological detection values ​​at each sampling time is then input into a fully connected layer and a sigmoid activation function for processing to obtain the risk assessment index of the target subject.

[0010] Preferably, the method for obtaining the temporal deterioration rate index of the target subject includes: , For the target subjects, it is an indicator of the rate of temporal deterioration; This represents the total number of sampling moments within the preset time window. For the target subjects at the sampling time Comprehensive physiological test values; The comprehensive physiological test values ​​at the initial sampling time within the time window; The total duration of the preset time window; This represents the sampling time sequence number within the time window.

[0011] It can quantify the recent accelerated deterioration trend of the subject's condition and effectively identify high-risk individuals in the acute fluctuation phase.

[0012] Preferably, obtaining the expected steady-state offset includes: , The expected steady-state offset of the target subject; The total follow-up period for the target subject's disease as specified in the clinical trial protocol, expressed in hours; For the target subjects, it is an indicator of the rate of temporal deterioration; It is a logarithmic function.

[0013] By using a logarithmic function to multiply the temporal deterioration rate index by the total duration of long-term clinical trial follow-up, it is possible to non-linearly amplify and explicitly expose potential safety hazards that may accumulate after patients enter a long trial period, providing a reliable quantitative pre-indicator for risk prediction in the mid-to-late stages of the trial.

[0014] Preferably, obtaining the degree of imbalance worsening in the allocation of target subjects to each candidate trial group includes: , The degree of imbalance worsening in assigning target subjects to the k-th candidate trial group; For the first The mean of the expected steady-state deviations of all subjects in each candidate trial group; To exclude the first The first candidate trial group The mean of the expected steady-state deviations of all subjects in each candidate trial group; The total number of candidate experimental groups; For the first The standard deviation of the expected steady-state deviation of all subjects in each candidate trial group; To prevent extremely small positive numbers with a denominator of zero; || represents the absolute value sign; This represents the expected steady-state offset of the target subject.

[0015] By quantitatively calculating the mean mutation caused by the simulated assignment of target subjects to candidate experimental groups, and introducing the standard deviation of the expected steady-state deviation of the group as a tolerance adjustment factor, the excessive concentration of high-risk patients with similar deterioration trends in the same experimental group can be effectively prevented.

[0016] Preferably, obtaining the final allocation probability score of the target subject for each candidate experimental group includes: , For the target subjects to the first Final allocation probability score for each candidate experimental group; Risk assessment index for target subjects; The degree of imbalance worsening in assigning target subjects to the k-th candidate trial group; The degree of imbalance worsening in assigning target subjects to the i-th candidate trial group; exp() represents an exponential function with the natural constant as the base; This represents the total number of candidate experimental groups.

[0017] Preferably, the grouping of target subjects based on their final allocation probability score for each candidate experimental group includes: The candidate experimental group corresponding to the maximum value of the final allocation probability score of the target subject to each candidate experimental group is taken as the target experimental group, and the target subject is assigned to the target experimental group.

[0018] Preferably, obtaining the normalized state detection data sequence for each dimension includes: A time-series coordinate axis is established based on a preset minimum time granularity. A linear interpolation algorithm is used to fill in missing values ​​in the state detection data sequence of each dimension. An isolated forest algorithm is introduced to identify and remove outliers. The min-max normalization method is used to map the state detection data sequence of the target subject in each dimension to a unified numerical scale range, so as to obtain the normalized state detection data sequence of the target subject in each dimension.

[0019] Secondly, the present invention provides a clinical subject grouping and processing system based on state detection, which adopts the following technical solution: A state-based clinical subject grouping system includes a processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the state-based clinical subject grouping method described above.

[0020] By adopting the above technical solution, the above-mentioned clinical subject grouping and processing method based on state detection is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0021] This invention has the following technical effects: By inputting normalized multi-dimensional state detection data sequences into a state feature extraction network to dynamically extract comprehensive physiological detection values, and combining the time decay term and the total duration of the trial follow-up period, the invention calculates the temporal deterioration rate index and the expected steady-state offset. This allows for in-depth analysis of the accelerated deterioration characteristics hidden behind fluctuations in conventional indicators, and accurate calculation of the cumulative risk deviation that may occur after subjects enter a long clinical trial period. Based on the expected steady-state offset, the invention calculates the degree of imbalance deterioration after subjects are simulated to join each group, and integrates it with risk assessment indicators to obtain a final allocation probability score for optimal grouping. This automatically punishes and accurately intercepts inappropriate allocation actions that lead to implicit risk imbalance, fundamentally ensuring a balanced distribution of disease conditions among candidate trial groups during long-term follow-up, and effectively preventing systematic errors and sudden safety events caused by excessive concentration of high-risk individuals. Attached Figure Description

[0022] Figure 1 This is a flowchart of the clinical subject grouping method based on state detection in an embodiment of the present invention; Figure 2 A graph showing the effect of data cleaning and normalization within a preset time window for the subjects; Figure 3 This is a comparison chart showing the imbalance in disease distribution among experimental groups during the clinical trial phase. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0024] This invention discloses a clinical subject grouping method based on state detection, referring to... Figure 1 This includes steps S1-S4: S1: Obtain the multi-dimensional state detection data sequence of the target subject within a preset time window, and perform time alignment and outlier cleaning to obtain a normalized state detection data sequence.

[0025] This embodiment is illustrated using a subject with cardiovascular disease as an example, with any subject with cardiovascular disease as the target subject; In this embodiment of the invention, a preset time window is used. Every hour, the system connects to the medical information system to obtain the target subject's vital signs data, laboratory test data, and symptom self-report data within the stated time window, forming a sequence of status detection data for the target subject in various dimensions.

[0026] The vital signs data include heart rate, blood pressure, or blood oxygen saturation monitored within the time window; the laboratory test data includes myocardial enzyme profiles, C-reactive protein, blood lipids, or brain natriuretic peptide (BNP or NT-proBNP) obtained from multiple blood samples within the time window; and the symptom self-report data includes subjective quantitative scores such as visual analog scale scores for pain, fatigue, or nausea levels, which are periodically recorded by the subject via a mobile terminal within the time window. Because the sampling frequencies of physiological detection items in different dimensions vary significantly, the system needs to perform time-series alignment on the multi-dimensional state detection data sequences. Specifically, the minimum time granularity is preset to one hour. The system establishes a unified time-series coordinate axis based on the minimum time granularity and uses a linear interpolation algorithm to fill in missing values ​​in the state detection data sequences of each dimension to maintain the continuity of the state sequences in the time dimension. At the same time, the system introduces an outlier detection mechanism based on the isolated forest algorithm to identify and remove outliers caused by sensor malfunctions or human input errors, ensuring the authenticity and reliability of the basic data.

[0027] After completing temporal alignment and outlier cleaning, the maximum-minimum normalization method is used to map the state detection data sequences of the target subjects in each dimension to a unified numerical scale range, thus obtaining the normalized state detection data sequences of the target subjects in each dimension.

[0028] S2: Input the normalized state detection data sequence of the target subject in each dimension into the pre-trained state feature extraction network to obtain the comprehensive physiological detection value of the target subject at each sampling time and the risk assessment index of the target subject; based on the comprehensive physiological detection value of the target subject at each sampling time, obtain the temporal deterioration rate index and expected steady-state offset of the target subject.

[0029] It should be noted that the physiological indicators of target subjects before formal enrollment often exhibit complex nonlinear characteristics. In particular, the recent sharp fluctuations in physiological indicators usually indicate that the physical condition is undergoing a hidden deterioration. If the group assessment relies solely on static values ​​at a single time point, it will be difficult to sensitively identify high-risk individuals in the acute fluctuation period, thus leading to a significant decline in the quality of grouping. Therefore, the present invention first requires inputting the target subject's multi-dimensional normalized state detection data sequence into a pre-trained state feature extraction network for feature extraction and fusion to obtain the target subject's comprehensive physiological detection value at each sampling time. The comprehensive physiological detection value can characterize the subject's overall physiological state at each sampling time. Then, based on the comprehensive physiological detection value, a temporal deterioration rate index of the target subject is constructed to capture the subject's recent disease deterioration trend, thereby providing a more clinically sensitive decision basis for subsequent randomization grouping.

[0030] In this embodiment of the invention, the normalized state detection data sequence of the target subject in each dimension is input into a pre-trained state feature extraction network for feature extraction and fusion. The state feature extraction network includes a long short-term memory network and an attention mechanism layer connected in sequence. Specifically, the normalized state detection data sequences of the target subject in each dimension are sequentially input into a long short-term memory network for processing to obtain the basic temporal feature matrix of the target subject. Then, the basic temporal feature matrix of the target subject is input into the attention mechanism layer to calculate the dynamic weight distribution vector of each dimension at each sampling time. The dynamic weight distribution vector of each dimension at each sampling time is weighted and summed with the normalized state detection data of the target subject in each dimension at each sampling time to output the comprehensive physiological detection value of the target subject at each sampling time. The time-series feature vector, composed of the comprehensive physiological test values ​​of the target subject at each sampling time, is sequentially input into a fully connected layer and a Sigmoid activation function. The fully connected layer performs linear combination and dimensionality reduction on the time-series feature vector, and the Sigmoid activation function nonlinearly compresses its output to the (0,1) interval to obtain the risk assessment index of the target subject.

[0031] It should be noted that the comprehensive physiological test values ​​characterize the overall physiological state of the subject at each sampling time, and the larger the value, the worse the state (the higher the degree of deterioration); the risk assessment index directly quantifies the overall risk level of the subject at the moment of randomization. Obtain the temporal deterioration rate index of the target subjects:

[0032] In the formula, For the target subjects, it is an indicator of the rate of temporal deterioration; This represents the total number of sampling moments within the preset time window. For the target subjects at the sampling time Comprehensive physiological test values; The comprehensive physiological test values ​​at the initial sampling time within the time window; The total duration of the preset time window; Represents the sampling time sequence number within the time window; As the sampling time t moves further away from the start of the time window, the time decay term... Increasing the value of gives higher weight to recent physiological changes, capturing recent trends in the subject's condition. If the target subject's physiological test value is significantly higher than the comprehensive physiological test value at the start of the sampling time within the time window, the relative deviation is considered higher. The larger the value, the larger the time decay term and the relative deviation become, and the greater the driving time deterioration rate index. Using a relative time scale As an exponential variable, it can perfectly fit the nonlinear clinical risk escalation model and strengthen the weight of recent real pathological deterioration. On the other hand, it constructs a bounded smooth mapping space, and the time decay term is strictly limited to the natural constant e, thereby completely avoiding the risk of numerical overflow caused by too many sampling times.

[0033] It should be noted that the temporal deterioration rate of the target subject is insufficient to comprehensively measure the profound impact that subject may have on the final outcome of the trial throughout the long clinical observation period. Existing clinical risk assessment methods often overlook the cumulative effect of the continuous evolution of the subject's condition. This can lead to an extremely dangerous situation where some subjects appear to be in a stable state upon enrollment, but the actual slope of their condition's evolution is already silently pointing towards a physiologically dangerous threshold. Once the trial enters its later stages, the target subject is highly susceptible to fatal safety risks. Therefore, this invention combines the temporal deterioration rate of the target subject with the observation period set for the entire clinical trial to obtain the expected steady-state offset of the target subject. Based on this expected steady-state offset, the true latent severity of the subject's condition after formally entering the trial can be reflected.

[0034] In this embodiment of the invention, the expected steady-state offset of the target subject is obtained:

[0035] In the formula, The expected steady-state offset of the target subject; The total follow-up period for the target subject's disease as specified in the clinical trial protocol, expressed in hours; For the target subjects, it is an indicator of the rate of temporal deterioration; It is a logarithmic function; The function can sensitively capture minor deterioration in the early stages of the disease, and can compress extremely high-risk values ​​to a reasonable range to prevent data overflow; As the independent variable deterioration rate index An increase in size leads to an increase in the product of itself and the observation period, thus increasing the size of the logarithmic function. The larger the value, the greater the expected steady-state offset. If the target subject grows rapidly, it could pose a fatal safety hazard during future trial periods.

[0036] S3: Based on the expected steady-state offset, obtain the degree of imbalance deterioration when the target subject is assigned to each candidate experimental group; based on the degree of imbalance deterioration and the risk assessment index of the target subject, obtain the final assignment probability score of the target subject to each candidate experimental group.

[0037] It should be noted that after obtaining the expected steady-state offset of the target subject, if the subject is still directly assigned to a candidate experimental group using the traditional allocation logic, it may disrupt the original disease distribution balance of the group. Furthermore, the traditional minimization allocation model cannot effectively identify inter-group differences based on the expected offset, and may easily concentrate subjects with similar deterioration trends in the same experimental group, thereby leading to uneven distribution of high-risk subjects among the groups. Therefore, this invention constructs an index for evaluating the degree of imbalance worsening between groups. This index quantitatively measures whether a significant numerical shift in the average expected steady-state offset will occur between a target subject and other groups after the target subject is simulatedly added to a candidate experimental group. A greater degree of imbalance worsening indicates a significant numerical shift, necessitating effective interception of inappropriate allocation actions that could exacerbate latent imbalances, thus ensuring the rationality and reliability of the clinical trial grouping process.

[0038] In this embodiment of the invention, by connecting to the clinical trial management system, the candidate trial groups corresponding to the clinical trial of the disease suffered by the target subject are obtained, and each candidate trial group contains subjects who have already been enrolled. Obtain the degree of imbalance worsening in the allocation of target subjects to each candidate trial group:

[0039] In the formula, The degree of imbalance worsening in assigning target subjects to the k-th candidate trial group; For the first The mean of the expected steady-state deviations of all subjects in each candidate trial group; To exclude the first The first candidate trial group The mean of the expected steady-state deviations of all subjects in each candidate trial group; The total number of candidate experimental groups; For the first The standard deviation of the expected steady-state deviation of all subjects in each candidate trial group; To prevent extremely small positive numbers with a denominator of zero, in this embodiment of the invention, a preset... In other embodiments, implementers may pre-set according to specific implementation conditions. The value of ; || represents the absolute value symbol; The larger the value, the more likely the target subject will be simulated and added to the first... After the first candidate experimental group, the first If there is a significant numerical abrupt change in the mean expected steady-state offset between the k-th candidate experimental group and other groups, then assigning the target subject to the k-th candidate experimental group would result in the mean of that group being significantly higher or lower than that of other groups, disrupting the balance between groups. Therefore, the k-th candidate experimental group... The candidate trial group should not be used as the target group for allocation of target subjects; When calculating the degree of imbalance deterioration, the following method is used: As the core denominator, its statistical theoretical basis comes from the standard score evaluation model in outlier detection. Because there are significant differences in the distribution of disease within different experimental groups at the initial stage of group establishment, the absolute difference cannot objectively measure the degree of inter-group imbalance caused by the addition of new individuals. Therefore, the standard score is introduced... As the denominator, it essentially constructs an index of intragroup disease tolerance: when the disease condition within a group is highly consistent (with a very small standard deviation), the denominator approaches zero, causing the calculation result to be significantly amplified. This imposes a strict penalty on any allocation action in that group that might disrupt the consistency of its disease condition, meaning that the group should not be used as an allocation target. Conversely, when the disease condition within a group itself has shown large fluctuations (with a large standard deviation), the denominator increases accordingly, and the degree of imbalance worsening is suppressed. This indicates that the group has a high tolerance for the inclusion of target subjects and is more suitable as an allocation target, thus achieving adaptive adjustment of intergroup imbalance assessment.

[0040] It should be noted that, in order to avoid implicit imbalance in the final dynamic randomization, the traditional allocation logic must be modified. By combining the degree of imbalance deterioration derived from time-series dynamics with the risk assessment index of the target subjects, the final allocation probability of the target subjects to each candidate experimental group is obtained, and the allocation tendency of the target subjects to each candidate experimental group is reconstructed.

[0041] In this embodiment of the invention, the target subject's response to the first... Final allocation probability score for each candidate experimental group:

[0042] In the formula, For the target subjects to the first Final allocation probability score for each candidate experimental group; Risk assessment index for target subjects; The degree of imbalance worsening in assigning target subjects to the k-th candidate trial group; The degree of imbalance worsening in assigning target subjects to the i-th candidate trial group; exp() represents an exponential function with the natural constant as the base; The total number of candidate experimental groups; As the imbalance worsens Risk assessment index The product term becomes larger, making The smaller the value, the more likely the target subject is to respond to the first... The smaller the final allocation probability score of each candidate experimental group, the less likely the target subjects will be to be assigned to experimental groups that would exacerbate the implicit imbalance, thus ensuring the balance and reliability of the grouping process.

[0043] S4: Group the target subjects based on their final allocation probability score for each candidate experimental group.

[0044] In this embodiment of the invention, the candidate experimental group corresponding to the maximum value of the final allocation probability score of the target subject to each candidate experimental group is taken as the target experimental group, and the target subject is assigned to the target experimental group.

[0045] Figure 2 The data cleaning and normalization effect diagram within the preset time window for the subjects shows the peaks (abnormal noise) and abruptly broken blank areas (missing values) in the original collected data. The normalized state detection data after processing by this invention uses linear interpolation to fill in missing values ​​and remove outliers, making the lines continuous and smooth, and mapping the values ​​to a uniform numerical scale range.

[0046] Figure 3This is a comparative chart of the imbalance in disease distribution among experimental groups during the clinical trial phase. The horizontal axis spans a clinical trial phase of up to twelve months. Existing technologies, due to blind randomization in the early stages, fail to identify potentially high-risk subjects, resulting in a severe imbalance in the disease distribution among groups in the later stages of the trial. In contrast, this invention can predict subjects who appear stable but are actually high-risk in advance. During the grouping phase, intelligent dynamic intervention is implemented through an imbalance deterioration index to evenly distribute such high-risk individuals among the experimental groups, maintaining a balanced and stable baseline of disease distribution among groups throughout the process, and ensuring comparability between groups and reliability of trial results during the long-term follow-up phase.

[0047] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for grouping clinical subjects based on state detection, characterized in that, include: The state detection data sequence of the target subject in each dimension within a preset time window is obtained. After time alignment, outlier cleaning and normalization, the normalized state detection data sequence of each dimension is obtained. The normalized state detection data sequence of the target subject in each dimension is input into the state feature extraction network to obtain the comprehensive physiological detection value and risk assessment index of the target subject at each sampling time. Based on the comprehensive physiological detection value, the temporal deterioration rate index of the target subject is obtained. Based on the temporal deterioration rate index and the total follow-up period specified in the clinical trial protocol for the target subject's disease, the expected steady-state offset is obtained. Based on the expected steady-state offset, the degree of imbalance worsening in the allocation of the target subjects to each candidate experimental group is obtained; based on the degree of imbalance worsening and the risk assessment index of the target subjects, the final allocation probability score of the target subjects to each candidate experimental group is obtained. The target subjects were grouped based on their final allocation probability score for each candidate trial group.

2. The clinical subject grouping method based on state detection according to claim 1, characterized in that, The process of acquiring state detection data sequences of the target subject in each dimension within a preset time window, followed by time alignment, outlier cleaning, and normalization, yields normalized state detection data sequences for each dimension, including: Preset time window Every hour, the system connects to the medical information system to obtain the target subject's vital signs data, laboratory test data, and symptom self-report data within the stated time window, forming a sequence of status detection data for the target subject in various dimensions.

3. The clinical subject grouping method based on state detection according to claim 1, characterized in that, The obtained comprehensive physiological test values ​​and risk assessment index of the target subject at each sampling time include: The normalized state detection data sequences of the target subject in each dimension are input into a long short-term memory network for processing to obtain the basic temporal feature matrix of the target subject. The basic temporal feature matrix is ​​then input into the attention mechanism layer to calculate the dynamic weight distribution vector of each dimension at each sampling time. The dynamic weight distribution vector is then weighted and summed with the normalized state detection data of each dimension at each sampling time to output the comprehensive physiological detection value of the target subject at each sampling time. The temporal feature vector composed of the comprehensive physiological detection values ​​at each sampling time is then input into a fully connected layer and a sigmoid activation function for processing to obtain the risk assessment index of the target subject.

4. The clinical subject grouping method based on state detection according to claim 1, characterized in that, The method for obtaining the temporal deterioration rate index of the target subject includes: , For the target subjects, it is an indicator of the rate of temporal deterioration; This represents the total number of sampling moments within the preset time window. For the target subjects at the sampling time Comprehensive physiological test values; The comprehensive physiological test values ​​at the initial sampling time within the time window; The total duration of the preset time window; This represents the sampling time sequence number within the time window.

5. The clinical subject grouping method based on state detection according to claim 1, characterized in that, The process of obtaining the expected steady-state offset includes: , The expected steady-state offset of the target subject; The total follow-up period for the target subject's disease as specified in the clinical trial protocol, expressed in hours; For the target subjects, it is an indicator of the rate of temporal deterioration; It is a logarithmic function.

6. The clinical subject grouping method based on state detection according to claim 1, characterized in that, The process of obtaining the degree of imbalance worsening in the allocation of target subjects to each candidate trial group includes: , The degree of imbalance worsening in assigning target subjects to the k-th candidate trial group; For the first The mean of the expected steady-state deviations of all subjects in each candidate trial group; To exclude the first The first candidate trial group The mean of the expected steady-state deviations of all subjects in each candidate trial group; The total number of candidate experimental groups; For the first The standard deviation of the expected steady-state deviation of all subjects in each candidate trial group; To prevent extremely small positive numbers with a denominator of zero; || represents the absolute value sign; This represents the expected steady-state offset of the target subject.

7. The clinical subject grouping method based on state detection according to claim 1, characterized in that, The process of obtaining the final allocation probability score of the target subject for each candidate experimental group includes: , For the target subjects to the first Final allocation probability score for each candidate experimental group; Risk assessment index for target subjects; The degree of imbalance worsening in assigning target subjects to the k-th candidate trial group; The degree of imbalance worsening in assigning target subjects to the i-th candidate trial group; exp() represents an exponential function with the natural constant as the base; This represents the total number of candidate experimental groups.

8. The clinical subject grouping method based on state detection according to claim 1, characterized in that, The grouping of target subjects based on their final allocation probability score for each candidate experimental group includes: The candidate experimental group corresponding to the maximum value of the final allocation probability score of the target subject to each candidate experimental group is taken as the target experimental group, and the target subject is assigned to the target experimental group.

9. The clinical subject grouping method based on state detection according to claim 1, characterized in that, The obtained normalized state detection data sequences for each dimension include: A time-series coordinate axis is established based on a preset minimum time granularity. A linear interpolation algorithm is used to fill in missing values ​​in the state detection data sequence of each dimension. An isolated forest algorithm is introduced to identify and remove outliers. The min-max normalization method is used to map the state detection data sequence of the target subject in each dimension to a unified numerical scale range, so as to obtain the normalized state detection data sequence of the target subject in each dimension.

10. A clinical subject grouping and processing system based on state detection, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement the clinical subject grouping method based on state detection according to any one of claims 1-9.

Images