A clinical trial whole-process quality closed-loop management and control system

Abstract

The application relates to the technical field of information system design services, and discloses a clinical trial whole-process quality closed-loop management and control system, which comprises a process atomization management and control module, a response sensitivity dynamic calibration module, a rule effectiveness drift self-compensation module, a behavior fingerprint verification module and a closed-loop feedback control module. The system maps clinical trial business flow data, determines the multi-dimensional compliance deviation degree of the multistage process management and control nodes, calibrates the response sensitivity of each node to quality control instructions, and corrects the judgment threshold of the checking operator accordingly. The system injects a logical micro-disturbance signal into the business data flow, extracts feedback characteristics, compares the behavior dynamics model, and outputs a authenticity verification result. The application establishes a quality management and control closed-loop system, uses the response sensitivity to reconcile the environmental heterogeneity contradiction, eliminates the quality gradient deviation, and realizes the self-evolution of the checking logic with the business evolution.

Description

Technical Field

[0001] This invention relates to a closed-loop quality control system for the entire clinical trial process, belonging to the field of information system design service technology. Background Technology

[0002] With the advancement of digitalization in clinical trials, collecting business data and applying logical verification rules has become a common practice. By using pre-defined compliance judgment logic to verify the input data stream, the accuracy and integrity of clinical trial data are ensured. In multi-center, large-scale clinical trial scenarios, the differences in operational standardization, researcher experience, and data acquisition terminal performance among research centers result in non-linear and discrete characteristics in the business data stream. Existing quality control architectures, when processing such asynchronous data streams, typically use fixed verification weights and uniform judgment thresholds, making it difficult to perceive the real-time compliance status characteristics of each research node. This leads to a mismatch between the feedback lag of verification actions and the dynamic evolution of the business environment. Because the system cannot adaptively identify the differences in quality absorption rates at each node, verification rules often cannot perform real-time compensation within the surface window where deviations occur, resulting in damage to the consistency boundary of clinical endpoint data.

[0003] Simply increasing the number of verification rules or blindly increasing the intensity of global verification not only increases the computational load of the system but also easily causes logical oscillations during system operation, leading to unnecessary blockages in business processes. This linear improvement approach cannot resolve the fundamental contradiction between static verification logic and multi-center dynamic business environments at the mechanism level. Simply improving the system architecture or increasing the overall management intensity is insufficient to resolve node compliance differences. Limitations at the control logic level are also prominent. For example, Chinese invention patent CN120032844A discloses a clinical trial quality control method and system that uses historical data to predict CRA manpower needs and performs static adaptation. This is an external resource overall scheduling approach that does not delve into the internal business flow to establish a logical perturbation and feedback compensation endogenous immune mechanism. Predictive management cannot sense the instantaneous response sensitivity of different execution entities to verification instructions. The rule effectiveness generates evaluation inertia drift in long-term trials and lacks self-calibration capabilities, failing to penetrate the false compliance barriers formed by human intervention.

[0004] Therefore, how to construct a control system with environmental logic stiffness identification and quality deviation closed-loop suppression functions to achieve real-time suppression of business data compliance residuals in heterogeneous environments has become the technical problem to be solved by this invention. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A closed-loop quality control system for the entire clinical trial process, the system comprising:

[0006] The process atomization control module is used to acquire clinical trial business flow data, deconstruct the clinical trial business flow data into multi-level process control nodes, and determine the multi-dimensional compliance deviation of each multi-level process control node based on preset compliance benchmarks.

[0007] The dynamic response sensitivity calibration module is used to monitor the execution response trajectory of each multi-level process control node to the quality control feedback instructions. By collecting the ratio of the change rate of multi-dimensional compliance deviation and the instruction adjustment weight of the quality control feedback instructions within a preset period, the dynamic response sensitivity of business execution is calculated.

[0008] The rule performance drift self-compensation module is used to define the judgment threshold for multi-dimensional compliance deviation and the verification operator for executing verification tasks. By statistically analyzing the mean offset of multi-dimensional compliance deviation, it identifies the performance drift characteristics of the judgment threshold relative to business evolution, and then corrects the value of the judgment threshold based on the sensitivity of dynamic response to business execution.

[0009] The behavioral fingerprint verification module is used to inject logical perturbation signals into clinical trial business flow data, extract the feedback features generated by the data link to the logical perturbation signals, compare the feedback features with the preset behavioral dynamics model, and output the data authenticity verification results.

[0010] The closed-loop feedback control module is used to output feedback gain adjustment instructions to each multi-level process control node based on the dynamic response sensitivity of business execution, the corrected judgment threshold, and the data authenticity verification results.

[0011] Preferably, the dynamic calibration module for response sensitivity is configured to perform the following steps: Step S1, obtain the multidimensional compliance deviation data sequence before and after executing the quality control feedback instruction; Step S2, calculate the convergence time of the step response of the multidimensional compliance deviation data sequence in the time domain, with the unit of measurement for the convergence time being ms; Step S3, normalize the ratio of the convergence time to the instruction adjustment weight to generate the dynamic response sensitivity of business execution; wherein, the dynamic response sensitivity of business execution is negatively correlated with the convergence time.

[0012] Preferably, the rule performance drift self-compensation module includes a residual statistics unit, which is used to monitor the mean offset of multidimensional compliance deviation data. When the mean offset shows a monotonically increasing trend within a preset monitoring window and the slope exceeds a preset slope threshold, the threshold is determined to have experienced performance drift.

[0013] Preferably, the rule performance drift self-compensation module increases the verification intensity of the corresponding high deviation interval by adjusting the weighted distribution function of the verification operator, and simultaneously reduces the computing resource allocation of the compliant interval.

[0014] Preferably, when injecting logical perturbation signals, the behavior fingerprint verification module adopts a non-predictive business field fine-tuning method, and the feedback features include data revision time, operation path backtracking features, and the logical consistency probability of field association correction.

[0015] Preferably, the behavioral fingerprint verification module calculates the Euclidean distance between the feedback features and the preset behavioral dynamics model. If the Euclidean distance is less than the preset anti-counterfeiting threshold, the output data authenticity verification result indicates that there is a risk of false consistency.

[0016] Preferably, the closed-loop feedback control module further includes a task flow load balancing unit, which is used to monitor the backlog rate of unprocessed tasks between two adjacent multi-level process control nodes.

[0017] Preferably, the task flow load balancing unit adjusts the task processing priority according to the following formula: ,in, For the first The processing priority weight of each task. The preset dimensionless adjustment factor, For the task The corresponding normalized absolute value of the multidimensional compliance deviation, This represents the real-time backlog of tasks waiting to be processed at this node. The sensitivity of dynamic response to the business execution corresponding to this node.

[0018] Preferably, the process atomization control module divides the clinical trial into the subject screening stage, the dosing follow-up stage, and the endpoint assessment stage, and sets the initial tolerance of the compliance benchmark according to the quality control accuracy requirements of each of the subject screening stage, the dosing follow-up stage, and the endpoint assessment stage.

[0019] Preferably, the system also includes a global evidence link solidification unit, which encapsulates feedback gain adjustment instructions, data authenticity verification results, and execution status of multi-level process control nodes into traceability metadata, and writes the traceability metadata into the distributed ledger in timestamp order.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] 1. In the entire process of clinical trial quality, by establishing a dynamic feedback mechanism for quality deviation perception and verification logic weight, real-time adaptation to the heterogeneous environment in multi-center clinical trials is achieved. This effectively smooths out the quality gradient caused by differences in personnel experience or equipment at different research nodes. By utilizing the residual characteristics generated at the output of each quality processing unit, the intervention gain is calculated through a closed-loop feedback controller, and the verification parameters for the next stage are reconstructed accordingly. This changes the traditional system's reliance on static, uniform threshold hard verification mode, enabling the system to complete deviation self-healing within the surface window of data flow, ensuring the consistency and stability of clinical endpoint data under complex concurrent conditions.

[0022] 2. By adopting a coordinated control strategy of logic stiffness identification and nonlinear gain correction, the system eliminates environmental oscillations caused by control commands under extreme conditions, ensuring the accurate allocation of quality control resources. By monitoring the mapping relationship between adjustment commands and changes in data thickness signals, the system can accurately calibrate the absorption characteristics of each research center and automatically switch to a small-step, high-frequency adjustment mode for high-resistance environments. This adaptive capability based on environmental perception avoids quality rebound caused by excessive correction in high-difficulty centers.

[0023] 3. A self-calibration algorithm based on the distribution correction of logic sensitivity coefficients is introduced. This system solves the evaluation inertia drift caused by the evolution of monitoring rules in long-term clinical trials. By real-time acquisition and analysis of residual distribution histograms, the system identifies the virtual performance decay caused by rule hardening and simulates the compensation process to dynamically adjust the distribution curve of the verification operator. This process can achieve self-evolution of the verification logic without modifying the underlying code, eliminating systematic errors caused by non-contact wear between the judgment benchmark and the actual situation on site, and maintaining the evaluation accuracy of the quality control system throughout its entire life cycle. Attached Figure Description

[0024] Figure 1 This is a diagram illustrating the core logic architecture and data flow principle of the system of this invention.

[0025] Figure 2 This is a diagram showing the functional module composition and hierarchical decomposition of the system of the present invention;

[0026] Figure 3 This is a diagram showing the interaction mechanism and closed-loop feedback topology of the various control neurons in this invention. Detailed Implementation

[0027] The technical features of this invention will be clearly and completely described below in conjunction with the technical solutions of this invention. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0028] This invention discloses a closed-loop quality control system for the entire clinical trial process, comprising a process atomization control module, a response sensitivity dynamic calibration module, a rule effectiveness drift self-compensation module, a behavioral fingerprint verification module, a closed-loop feedback control module, and a global evidence link solidification unit cascaded together. Through a preset data interaction interface, it achieves real-time processing and quality feedback of clinical trial business flow data. The system deconstructs clinical business logic into measurable multi-level process control nodes through the process atomization control module, utilizes the response sensitivity dynamic calibration module and the rule effectiveness drift self-compensation module to correct parameters for environmental differences and rule aging issues, and then uses the behavioral fingerprint verification module to verify data authenticity. Upon verification, the closed-loop feedback control module outputs a feedback gain adjustment command. To address the nonlinear and discrete characteristics of workflow data in large-scale, multi-center clinical trials, the process atomization control module performs workflow deconstruction and compliance evaluation. This module acquires raw workflow data from the subject screening, dosing follow-up, and endpoint assessment phases via a data acquisition interface, and deconstructs the data flow into multi-level workflow control nodes composed of a series of atomic operations. The dosing follow-up phase is deconstructed into drug preparation, dosing time verification, subject vital sign collection, and adverse event recording nodes. The module calculates the multidimensional compliance deviation of each node based on preset compliance benchmarks. The multidimensional compliance deviation value is a weighted composite of three physical components: the first component is the absolute value of the deviation of the business operation time (in milliseconds) from the standard time window; the second component is the offset ratio of the entered value from the boundary of the allowable fluctuation range; and the third component is the logical XOR result of the business step execution order and the preset logical path (0 for logical consistency, 1 for inconsistency). The system captures the above data in real time at a sampling frequency of 10Hz, multiplies the three components by weight coefficients of 0.4, 0.3, and 0.3 respectively, and then sums them to convert the complex business state into a normalized multidimensional compliance deviation value. The compliance status of a node is characterized by comparing the semantic Euclidean distance between the business feature vector output by the node and the standard compliance template vector. When the subject follow-up time deviation exceeds a preset window and laboratory records are missing, the system calculates... As the numerical value increases, the system uses this atomized deconstruction method to capture the deviation characteristics of each research center within the data acquisition time window, providing a basis for subsequent quality deviation mitigation.

[0029] To address the technical challenge of varying responsiveness to quality control directives across different clinical research centers due to differences in personnel experience or equipment performance, the system employs a dynamic response sensitivity calibration module to monitor the execution response trajectory of each node. This module executes the following calibration procedure, collecting multidimensional compliance deviations within a preset period. The system calculates the convergence time of the step response of the multidimensional compliance deviation data sequence before and after the execution of the quality control feedback instruction, based on the proportional relationship between the rate of change and the instruction adjustment weight. The unit of measurement for this convergence time is set to milliseconds (ms). The specific convergence judgment logic is as follows: The system continuously monitors the multidimensional compliance deviation value in 10ms increments. When the value enters the error band of ±3% of the target setpoint, and the maximum fluctuation of the value is less than 0.1% in the subsequent 10 consecutive sampling periods, the controller determines that the signal has reached steady-state convergence and records the cumulative time from the instruction issuance time to the start of this steady-state as the convergence time. The system normalizes the ratio of this convergence time to the instruction adjustment weight to generate a business execution dynamic response sensitivity used to characterize the response efficiency of this node. ,in It shows a negative correlation with the convergence time. If the system sends a correction instruction to the node to increase the weight by 5 units, its convergence time will be negatively affected. The former tends to stabilize within 200ms, while the other node takes 800ms to converge after receiving the same instruction. Therefore, the former's... This is higher than the latter, thus the system quantifies the logical stiffness under different research environments, avoiding system oscillations caused by excessively rapid adjustments or the accumulation of deviations caused by slow corrections. To address the issue of declining evaluation effectiveness of quality control rules due to business evolution in long-term clinical trials, the rule effectiveness drift self-compensation module executes the calibration procedure of the verification operator. The verification operator is a set of logical verification tasks pre-set for the business logic of clinical trials. It contains a data mapping function for capturing real-time values ​​of specific control nodes, a benchmark judgment module with built-in compliance thresholds, and a weight adjustment factor that determines the execution frequency. The residual statistics unit monitors multidimensional compliance deviations. The mean deviation of the data is considered. When this mean deviation shows a monotonically increasing trend within a preset 24-hour monitoring window, and the slope of its regression curve exceeds a preset slope threshold of 0.15, the system determines that the performance drift of the verification operator for that node has occurred. This 0.15 slope threshold is an engineering boundary point derived from least squares regression analysis of 35,000 sets of deviation data generated by eight research centers of different sizes over a cumulative 1200-hour operating cycle. When the concurrent task load increases from 10 times per second to 100 times per second, if the rising slope of the mean deviation crosses 0.15, the missed detection rate of the system's verification operator will surge from 2.5% to 18.2%. Therefore, this value is used as a deterministic criterion for triggering the self-compensation mechanism. To mitigate this rule deviation, the system is based on the sensitivity of dynamic response to business execution. The system corrects the judgment threshold by adjusting the weighted distribution function of the verification operator. The weighted distribution function adopts a method based on... The modified Gaussian distribution offset model defines the verification intensity weights. Deviation from Multidimensional Compliance The mapping relationship is given by the following formula: ,in, For the corresponding deviation The following are the weight allocation values ​​for the verification operators; The mean offset of the multidimensional compliance deviation monitored by the residual statistical unit; This is a width adjustment factor for the weight distribution, and its value is affected by the sensitivity of the dynamic response of business execution. The reverse correction; To enhance the sensitivity of dynamic response to business execution at this node, the verification weight for high deviation ranges is increased, while the computing resource allocation for compliant ranges is simultaneously reduced. The resource allocation reduction mechanism follows the principle of total conservation, specifically by setting the total computing resource capacity of the system to a constant. Resource allocation within the compliance period during each verification cycle. Dynamic stripping is performed according to the following formula: ,in, The allocation of computing resources within the compliant range; The total computing resource capacity weight is preset for the system; The preset performance drift trigger threshold is used; by weighting the high deviation interval... The system calculates points and deducts redundant resources from compliant areas, enabling precise transfer of computing resources from low-risk to high-risk regions. When it detects that data distribution is converging towards a threshold boundary due to changes in subject group characteristics, the system automatically increases the verification intensity of that node by 15% and passes the verification. Factors are corrected, specifically, through The steps for factor correction are as follows: Step S101, obtain the initial benchmark judgment threshold preset by the rule performance drift self-compensation module. Step S102: Call the dynamic response sensitivity of the service execution output by the dynamic response sensitivity calibration module. Step S103: Calculate the corrected dynamic judgment threshold. The calculation formula is as follows: ,in, This is the corrected final judgment threshold; The threshold is used as the benchmark for judgment; Sensitivity to dynamic responses in business execution; For the mean offset of the multidimensional compliance deviation data within the monitoring window, step S104, will... Feedback is sent to the judgment module of the verification operator, replacing the original fixed threshold. Through the above steps, the node response sensitivity is... When it decreases (representing poor environmental absorption), Automatic tightening smooths out compliance residuals during data flow, maintaining the evaluation accuracy of the quality control system throughout its entire lifecycle.

[0030] When determining the threshold for the verification operator, the rule performance drift self-compensation module uses the residual statistics unit to analyze the multidimensional compliance deviation. Sliding window monitoring selects a preset 24-hour monitoring window. Least squares regression analysis is used to analyze the mean shift within the window, yielding a regression curve slope reflecting the data evolution trend. When the slope exceeds a preset slope threshold of 0.15, it is determined that the evaluation effectiveness of the verification operator has shifted, and the dynamic response sensitivity of the business execution output from the dynamic response sensitivity calibration module is invoked. The judgment threshold is corrected by adjusting the weighted distribution function of the verification operator, based on the aforementioned weight allocation formula. and resource stripping formula While maintaining a constant overall computing resource allocation, the system linearly increases the weight of verification intensity in high deviation intervals, while simultaneously reducing computing resource usage in compliant intervals proportionally. This allows the system to adjust the intervention gain based on the absorption rate characteristics of different research centers, thus enabling multi-dimensional compliance deviation... Within an 800ms feedback cycle, the system returns to the preset compliance balance window, achieving real-time mitigation of compliance residuals in business data under heterogeneous environments. Addressing the potential risk of false consistency in clinical trials, the behavioral fingerprint verification module performs data authenticity checks. This module injects standardized logical perturbation signals into the clinical trial business flow data, generated using a non-predictive business field fine-tuning method. It randomly triggers review requests for non-critical fields or logical jumps in the data entry interface. By modifying the logical branches of the terminal rendering engine, without altering the medical meaning of the data, it inserts a dynamic random delay of 40ms to 120ms into page jump instructions, or performs a bit flip of 1 bit in a non-mandatory text comment box. The module is pre-loaded with 85... The whitelist mapping table for non-critical business fields ensures that perturbation actions are only executed randomly within the whitelist range, preventing backend database transaction rollbacks or logical deadlocks due to perturbations. The system extracts feedback characteristics of the data link generated by the perturbation signal, including data revision time, operation path backtracking characteristics, and logical consistency probability of field association correction. The system calculates the Euclidean distance between the above feedback characteristics and the preset behavioral dynamics model to output the data authenticity verification result. If the calculated Euclidean distance is less than the preset anti-counterfeiting threshold of 0.05, it indicates that the data response trajectory exhibits unnatural human correction characteristics, i.e., it is judged to have a risk of false consistency. This module identifies abnormal data flows through process verification, enhancing the evidentiary strength of clinical data.

[0031] The behavioral fingerprint verification module collects the coordinate sequence and time interval of the terminal interaction interface, and extracts a multi-dimensional feature vector composed of operation trajectory curvature, field switching time, and backtracking correction frequency. The detailed construction process of this behavioral dynamics model is as follows: The system performs data cleaning and benchmark collection, retrieving 2000 original operation records that have been manually verified as compliant by the research center within 45 natural days from the historical compliant sample set stored in the distributed ledger as the training sample set; it performs feature space dimensionality reduction, using principal component analysis to reduce the dimensionality of the feature vector space containing more than 100 operation units, selecting the top dimensionality with a cumulative contribution rate greater than 85%. Each principal component constitutes a characteristic subspace. The centroid determination procedure is executed, and the arithmetic mean of the characteristic subspace is calculated to determine the characteristic centroid of the research center. The model introduces a time sliding window mechanism. When the number of stored real samples reaches 1000 and the time span is greater than 30 calendar days, the system recalculates the centroid position to achieve adaptive iteration of model parameters. is the number of features for dimensionality reduction, and is a positive integer. Let be the feature centroid, and be a dimensionless parameter. Calculate the relationship between the current fingerprint vector and the feature centroid. The semantic Euclidean distance between them is calculated using the following specific steps to verify the data authenticity: Step 1, Component difference calculation: Calculate the fingerprint vector generated by the current operation. Along each principal component dimension and the eigencentroid The difference of coordinates Step two, sum the distances according to the formula. Calculate the current fingerprint vector and feature centroid. The semantic Euclidean distance between them, where The calculated semantic Euclidean distance, For the current fingerprint vector at the th The components on each principal component The components of the feature centroid in the corresponding dimension. The number of dimensionality reduction features selected; Step 3, dynamic comparison, using semantic Euclidean distance With dynamic anti-counterfeiting threshold Logical comparison, if If the data response trajectory exhibits non-natural, artificially corrected characteristics, the anti-counterfeiting threshold is set. The calculation formula is as follows: ,in The proportionality constant is 0.05, which is a dimensionless parameter. The standard deviation of fingerprint vectors from the historical sample set during the baseline period, in milliseconds. To assess the multidimensional compliance deviation of current control nodes and determine whether the data response trajectory exhibits non-natural, artificially corrected characteristics, data authenticity verification is completed without altering the semantics of clinical trial business. The specific behavioral dynamics model is implemented by constructing a 3D feature space, with coordinate axes representing: the pixel rate of cursor movement on the terminal interface, the millisecond interval between key presses and releases, and the switching time M between unrelated fields. The core architecture is defined as a spatial distribution based on an n-dimensional feature vector. In this embodiment, the behavioral dynamics model constructs a 3D feature space vector V. fp =[f speed ,f interval ,f switch This is achieved by [the following], with the coordinate axes being: the pixel movement rate f of the cursor on the terminal interface. speed The time interval (finterval) between key presses and releases, and the switching time (f) between unrelated fields. switch This model not only includes the feature centroid C ref It also includes the standard deviation σ of the baseline period sample. ref The established probability envelope range is determined by retrieving 2000 manually verified compliant operation records from the past 45 natural days, calculating the arithmetic mean of these three physical quantities in space to establish the feature centroid coordinates, and setting a spherical judgment envelope with a radius of 0.05 units around this centroid. If the current operation fingerprint vector falls outside this spherical envelope, it is determined that there is a risk of human intervention. During the deviation smoothing process at each node, the closed-loop feedback control module handles the task backlog problem in multi-center collaboration through the task flow load balancing unit. The task flow load balancing unit monitors the backlog rate of unprocessed tasks between two adjacent multi-level process control nodes and adjusts the processing order of each task according to the priority adjustment formula, which is as follows: ,in, For the first The processing priority weight of each task is a dimensionless parameter. The preset dimensionless adjustment factor is 0.6. For the task The normalized absolute value of the corresponding multidimensional compliance deviation is a dimensionless parameter. This represents the real-time backlog of tasks to be processed at this node, expressed in units. This is the dynamic response sensitivity of the business execution corresponding to this node, which is a dimensionless parameter.

[0032] Through this calculation procedure, the system initiates task priority reordering when task backlog deviates from the preset balance window. When the backlog rate of backend processing nodes exceeds 80%, the system reduces the output traffic of high-sensitivity frontend nodes and promotes high-risk tasks. Prioritizing larger tasks achieves load balancing and prevents performance degradation caused by localized congestion. To ensure the legal traceability of the control process, the global evidence chain solidification unit intercepts the runtime metadata of each atomic control node in real time. This metadata is composed of the real-time compliance deviation vector of each node, feedback gain adjustment instructions, and behavioral fingerprint verification results. The system calls a secure hash algorithm to iteratively calculate the current runtime metadata and the root hash value of the previous block in the distributed ledger to generate a traceability feature digest for the current node. The calculation formula is as follows: ;in, The source feature summary generated for the current node is a dimensionless parameter; The default secure hash function; For the first The runtime metadata of each control node; The feature summary value, which is a dimensionless parameter, is fixed in the distributed ledger by the previous node. This procedure establishes temporal causal constraints between various quality control actions through chained hash anchoring. Once a unit bit mutation occurs in the historical record of a certain node, the resulting feature residual will cause the hash verification of the entire evidence chain to fail, thus realizing anti-tampering protection for the quality control behavior of the entire clinical trial process. When the system detects that the current traceability feature summary does not match the hash value of the previous block, the closed-loop feedback control module will execute the interruption logic, forcibly set the value of the feedback gain adjustment instruction sent to the corresponding process control node to 0, and simultaneously reduce the task processing priority to the lowest level. The logic block can only be released and the feedback loop reactivated after manual intervention to complete the legality reset of the distributed ledger.

[0033] Example 1: In a multi-regional, multi-center clinical trial environment comprising 12 research centers, the varying depth of understanding of operating procedures among clinical coordinators at each center led to nonlinear time lag fluctuations in the dosing time records during the subject screening phase, ranging from 120ms to 450ms relative to the preset window. This quality gradient bias, caused by environmental heterogeneity, resulted in a mismatch between the response lag of control actions and the logical rigidity of the environment when applying traditional static verification methods. Corrective actions triggered quality rebound and exacerbated data fluctuations, ultimately damaging the consistency boundary of clinical endpoint data. To address these application-layer challenges, the process atomization control module acquires real-time business flow data during the follow-up phase and deconstructs this data into multi-level process control nodes consisting of drug preparation nodes, dosing verification nodes, and vital sign collection nodes. The response sensitivity dynamic calibration module monitors the execution response trajectory of each research center when executing quality control feedback instructions. The system calculates the multi-dimensional compliance deviation before and after executing the quality control feedback instructions. The convergence time of the step response in the time domain is used to calibrate the sensitivity of the dynamic response of business execution corresponding to the high time delay research center. The rule-based performance drift self-compensation module is based on The quantified value is used to correct the judgment threshold of the verification operator. When the slope of the regression curve of the multidimensional compliance deviation mean of a research center exceeds 0.15 within the 24-hour monitoring window, the system adjusts the weighted distribution function of the verification operator to increase the verification weight of the corresponding high deviation interval without modifying the underlying code. As a regulating factor for the intensity of verification, the system can automatically adjust the intervention gain according to the absorption characteristics of different research centers, thus eliminating the contradiction between excessive intervention in highly sensitive centers and insufficient correction in low-sensitive centers caused by uniform verification standards.

[0034] Simultaneously, the behavioral fingerprint verification module injects standardized logic perturbation signals for flaw detection into the data link. By collecting the spatial distance between the data revision time and the preset behavioral dynamics model, it performs authenticity penetration verification. The system transforms data authenticity determination into dynamic causal verification of operational feedback characteristics. Under the coordinated action of the closed-loop feedback control module, the system adjusts according to the priority formula. Dynamically rearrange backlogged tasks, utilizing smaller... The value is increased to prioritize the quality control processing of research centers with slow response times, ensuring that high-risk tasks receive priority in resource allocation. Under the continuous adjustment of this feedback control loop, all research centers... The data returns to the preset compliance equilibrium window within an 800ms period, and the quality data throughout the entire process shows convergence.

[0035] Example 2: In a clinical data management system validation platform comprising 15 virtual research centers with heterogeneous logical response characteristics, the system acquires data from business data sequences generated by a discrete event simulator conforming to standard clinical trial data exchange standards. This simulates data flow under multi-center concurrent conditions. The experiment aims to verify the response sensitivity dynamic calibration module's response to multidimensional compliance deviations in a Gaussian white noise environment with a superimposed 20dB signal-to-noise ratio. The ability to suppress, priority adjustment factor The settings involve a trade-off between the rate of deviation correction and flow stability, when When the value approaches 1, the system prioritizes the immediate processing of high-risk tasks, but this can lead to a temporary backlog of tasks at processing nodes. Conversely, a value closer to 1 helps maintain stable traffic flow between nodes. Through sensitivity analysis under various operating conditions, the system demonstrated its performance in this experiment. The value is set to a dimensionless constant of 0.6 to ensure the efficiency of error correction while keeping the overall task backlog rate of the system below 10%.

[0036] The specific experimental process was initiated by injecting business flow data with different time delay characteristics into research centers D1 to D8. D1 was a simulated high-execution center with an instruction absorption rate of 0.95, while D8 was a low-response center with an instruction absorption rate of only 0.25. Table 1 compares the quality control performance of different experimental groups under noisy conditions. Table 1 records the data comparison of each center before and after applying the present invention. The initial deviation reflects the initial data state including 15% human input jitter error. Observing the data in Table 1, it can be seen that when the control group uses a uniform judgment threshold without enabling dynamic response sensitivity correction, the final deviation of the slow-response centers D5 to D8 shows a divergent trend due to feedback time delay. In contrast, the experimental group used the dynamic response sensitivity calibration module to calibrate the dynamic response sensitivity of business execution. The decision weights of the verification operators have been adjusted. Data shows that, with... The deviation was reduced from 0.92 to 0.38, and the final deviation of the experimental group remained within a stable window of 0.05, proving that the system can smooth out the heterogeneous quality gradient by quantifying the environmental logic stiffness.

[0037] Table 1: Example of Comparison of Quality Control Performance of Different Test Sample Groups

[0038] Further self-compensation experiments on the performance drift of the verification operator revealed that when the multidimensional compliance deviation... When the slope of the regression curve for the mean deviation fluctuates between 0.05 and 0.12 within 24 hours, the compliance residual output by the system remains stable. However, once the slope crosses the performance inflection point of 0.15, the mean deviation of the control group without activated rule performance drift self-compensation module shows a non-linear increase. At this point, the experimental group with activated compensation mechanism reduces the deviation growth rate by 72% by adjusting the weighted distribution function of the verification operator. This phenomenon proves that the slope threshold of 0.15 is an effective engineering boundary point for judging the aging of quality control rules, preventing the accumulation of systematic errors due to evaluation inertia in long-term test environments. Finally, the test data shows that under the interference of non-linear time delay fluctuations of 120ms to 800ms, this solution will reduce the multi-dimensional compliance deviation across the entire line. It is controlled within a preset threshold below 0.06.

[0039] Example 3: This example combines Figures 1 to 3 Describe a closed-loop quality control system for the entire clinical trial process, such as... Figure 1As shown, the logical architecture starts with input: clinical trial business flow data. This data flow enters the process atomic control module to perform processing operations such as deconstructing business nodes and determining multi-dimensional compliance deviations. The data flow is divided into multiple transmission paths. One path enters the response sensitivity dynamic calibration module to monitor the execution response trajectory and calculate the dynamic response sensitivity, and generates a signal based on sensitivity correction to be transmitted to the rule performance drift self-compensation module. The other path directly enters the rule performance drift self-compensation module to complete the identification of performance drift characteristics and correct the judgment threshold of the verification operator. The main data flow passes through the behavior fingerprint verification module to perform the injection of logical perturbation signals and output the authenticity verification results. The processing results of the above modules finally converge to the closed-loop feedback control module, which performs comprehensive sensitivity and threshold analysis and task flow load balancing control, and finally generates and issues the output: feedback gain adjustment command.

[0040] like Figure 2 As shown, the system's functional composition is further deconstructed into six functional branches pointing to a closed-loop quality control system for the entire clinical trial process. These include: a process atomization control branch (comprising business flow deconstruction into control nodes and calculation of multi-dimensional compliance deviation); a response sensitivity calibration branch (monitoring execution response trajectories and calculating step response convergence time); a rule performance drift compensation branch (identifying threshold performance drift and using sensitivity-based correction operators); a behavioral fingerprint verification branch (compiling a comparison dynamics model and injecting logical perturbation signals); a closed-loop feedback control branch (task flow load balancing and output feedback gain instructions); and an evidence link solidification branch (ensuring data immutability by writing to a distributed ledger and encapsulating traceability metadata). Figure 3 As shown, the system's functional units present a neuron-like interactive topology. The clinical trial business data flow serves as the input source, flowing sequentially through the process atomization control neuron and the behavior fingerprint verification neuron that receives logical perturbation injection, ultimately reaching the closed-loop feedback control center. Simultaneously, the sensitivity calibration neuron processes the data flow in parallel and provides input to the control center. The rule efficiency compensation neuron establishes a connection with the control center after receiving the weight correction signal. All interactions are ultimately processed by the control center, which generates feedback gain adjustment commands and feeds them back to the front-end business data flow through the closed-loop adjustment loop.

[0041] Example 4: In the long-term, multi-center clinical trial environment for anti-tumor drugs, subject data entry faces heterogeneity from multiple research centers. Due to the potential risk of some centers automatically filling in data using entry scripts or maliciously tampering with it, the logical relationships between business fields exhibit excessive consistency beyond natural behavioral distribution. This behavioral forgery masks physiological fluctuations and operational randomness during data collection, making it difficult for the system to identify the authenticity of the data source and posing a risk to the consistency determination of clinical research endpoints. The behavioral fingerprint verification module is used to perform non-predictive logical perturbation based on bitmask mapping. The system calls a pseudo-random number generator to select from a preset set of non-critical business fields. The system identifies target locations and adjusts the jump sequence of these locations on the input interface based on the bit-flipping matrix. This allows the terminal interface to generate unpredictable logical path fluctuations without altering the business semantics. The system simultaneously collects the feedback time series of the input personnel to these fluctuation signals and the movement trajectory of the cursor in the two-dimensional coordinate system. Using principal component analysis, the system maps these features into fingerprint vectors that characterize the dynamic features of the input behavior. By calling the historical legal sample set stored in the global evidence link solidification unit, the system calculates the semantic Euclidean distance between the current fingerprint vector and the feature centroid of the corresponding research center's behavioral dynamics model. The model is based on the probability density center distribution generated by the operation trajectory manually verified by the research center during the baseline period.

[0042] The system executes an automatic calibration procedure for anti-counterfeiting thresholds to eliminate subjectivity in the judgment process. The calculation formula is as follows: ;in, The anti-counterfeiting threshold is a dimensionless parameter. The proportionality coefficient is 0.05, which is a dimensionless parameter; represents the standard deviation of fingerprint vectors in the historical sample set during the baseline period, in milliseconds. This represents the multidimensional compliance deviation of the current control node, which is a dimensionless parameter; the formula utilizes... A dynamic correlation was established between anti-counterfeiting sensitivity and compliance risk. When the multidimensional compliance deviation of a certain node increases, the system reduces... The value of is chosen to improve the sensitivity of capturing behavioral anomalies. When the semantic Euclidean distance of the current fingerprint vector is greater than 1, the sensitivity is improved. At that time, the closed-loop feedback control module reduces the sensitivity of the dynamic response of the business execution corresponding to the task. The weights are assigned to prioritize tasks. During the calculation process, it is automatically downgraded and marked as evidence pending verification, thus mitigating quality deviations at their source.

[0043] Example 5: In the initial deployment of a distributed management network including newly connected clinical research centers, the system initiates an initial benchmark calibration procedure to establish the logical evaluation origin for this environment. The process atomic management module is in monitoring mode for the first 14 consecutive working days after access. By collecting 500 business entry records that have been manually cross-checked and confirmed as compliant as a benchmark sample, the initial compliance template vector and multidimensional compliance deviation of the research center are calculated. The environmental background noise value is used as the basis for dynamic calibration of response sensitivity. The module sends a simulated quality control command containing 10 test cases to the terminal. By measuring the variance change of the time series feedback from the terminal, the dynamic response sensitivity of the center under standard operating conditions is determined. The calibration process uses a differential algorithm to deduct the fixed time delay caused by hardware communication delay, thereby establishing the initial logic stiffness of the closed-loop feedback control loop connected to the center.

[0044] When constructing the sample space, the behavioral fingerprint verification module performs desensitization processing on the retrieved historical operation records using the global evidence link solidification unit. The system uses a hash mapping algorithm to convert the user's identity identifier into a unique anonymized code, and uses principal component analysis to reduce the dimensionality of the feature vector space covering more than 100 operation units, selecting the top features with a cumulative contribution rate greater than 85%. The principal components constitute the feature subspace, and the feature centroids are determined based on the arithmetic mean logic. ,in, The characteristic centroid is a dimensionless parameter. The selected number of dimensionality reduction features is a dimensionless parameter. The centroid is dynamically iterated based on the time sliding window mechanism. When the number of stored real samples reaches 1,000 and the time span is greater than 30 natural days, the system recalculates the centroid position to smooth out the deviation caused by the evolution of input habits, thereby establishing the evaluation benchmark for data authenticity flaw detection.

[0045] Example 6: During the deployment phase of a multi-center clinical trial, the system executes pre-calibration procedures. The process atomization control module retrieves quality control indicators from the clinical monitoring plan through the data acquisition interface, converting subject compliance, data entry timeliness, and adverse event reporting delays into initial compliance template vectors. The system collects multidimensional compliance deviation data over 100 consecutive closed-loop feedback cycles in a simulated training environment, and determines the environmental baseline residual by calculating the variance distribution of the data during operation. Furthermore, the differential algorithm is used to deduct systematic deviations caused by network latency, thereby calibrating the perception accuracy of each process atomic control node.

[0046] The system injects a step correction signal into the terminal to measure the multidimensional compliance deviation. Return to Response time for 90% of the width within the range And determine the critical sensitivity threshold based on the critical sensitivity formula. The critical sensitivity formula is as follows: ;in, The critical sensitivity threshold is a dimensionless parameter. The adjustment coefficient is 0.45, which is a dimensionless parameter; The peak deviation after the injection of a step signal is a dimensionless parameter. The maximum allowed backlog of tasks, expressed in units of [number]; the sensitivity of dynamic response to business execution obtained from online monitoring. Three consecutive sampling points below At this time, the closed-loop feedback control module automatically reduces the proportional gain of the feedback loop and outputs a logic blockage warning work order to the control terminal, guiding the system to switch to manual verification mode to smooth the quality gradient divergence caused by the decline in environmental responsiveness.

[0047] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A closed-loop quality control system for the entire clinical trial process, characterized in that, The system includes: The process atomization control module is used to acquire clinical trial business flow data, deconstruct the clinical trial business flow data into multi-level process control nodes, and determine the multi-dimensional compliance deviation of each multi-level process control node based on preset compliance benchmarks. The dynamic response sensitivity calibration module is used to monitor the execution response trajectory of each multi-level process control node to the quality control feedback instructions. By collecting the ratio of the change rate of multi-dimensional compliance deviation and the instruction adjustment weight of the quality control feedback instructions within a preset period, the dynamic response sensitivity of business execution is calculated. The rule performance drift self-compensation module is used to define the judgment threshold for multi-dimensional compliance deviation and the verification operator for executing verification tasks. By statistically analyzing the mean offset of multi-dimensional compliance deviation, it identifies the performance drift characteristics of the judgment threshold relative to business evolution, and then corrects the value of the judgment threshold based on the sensitivity of dynamic response to business execution. The behavioral fingerprint verification module is used to inject logical perturbation signals into clinical trial business flow data, extract the feedback features generated by the data link to the logical perturbation signals, compare the feedback features with the preset behavioral dynamics model, and output the data authenticity verification results. The closed-loop feedback control module is used to output feedback gain adjustment instructions to each multi-level process control node based on the dynamic response sensitivity of business execution, the corrected judgment threshold, and the data authenticity verification results.

2. The clinical trial end-to-end quality closed-loop control system according to claim 1, characterized in that, The dynamic response sensitivity calibration module is configured to perform the following steps: Step S1, obtain the multidimensional compliance deviation data sequence before and after executing the quality control feedback instruction; Step S2, calculate the convergence time of the step response of the multidimensional compliance deviation data sequence in the time domain, with the unit of measurement for the convergence time being ms; Step S3, normalize the ratio of the convergence time to the instruction adjustment weight to generate the dynamic response sensitivity of business execution; wherein, the dynamic response sensitivity of business execution is negatively correlated with the convergence time.

3. The clinical trial end-to-end quality closed-loop control system according to claim 1, characterized in that, The rule performance drift self-compensation module includes a residual statistics unit, which is used to monitor the mean deviation of multidimensional compliance deviation data. When the mean deviation shows a monotonically increasing trend within the preset monitoring window and the slope exceeds the preset slope threshold, it is determined that the threshold has experienced performance drift.

4. The clinical trial end-to-end quality closed-loop control system according to claim 1, characterized in that, The rule performance drift self-compensation module increases the verification intensity of the corresponding high deviation range by adjusting the weighted distribution function of the verification operator, and simultaneously reduces the computing resource allocation of the compliant range.

5. The clinical trial end-to-end quality closed-loop control system according to claim 1, characterized in that, When injecting logical perturbation signals, the behavioral fingerprint verification module adopts a non-predictive business field fine-tuning method. The feedback features include data revision time, operation path backtracking features, and the logical consistency probability of field association correction.

6. The clinical trial end-to-end quality closed-loop control system according to claim 1, characterized in that, The behavioral fingerprint verification module calculates the Euclidean distance between the feedback features and the preset behavioral dynamics model. If the Euclidean distance is less than the preset anti-counterfeiting threshold, the output data authenticity verification result indicates that there is a risk of false consistency.

7. The clinical trial end-to-end quality closed-loop control system according to claim 1, characterized in that, The closed-loop feedback control module also includes a task flow load balancing unit, which is used to monitor the backlog rate of unprocessed tasks between two adjacent multi-level process control nodes.

8. The clinical trial end-to-end quality closed-loop control system according to claim 7, characterized in that, The task flow load balancing unit adjusts task processing priority according to the following formula: ,in, For the first The processing priority weight of each task. The preset dimensionless adjustment factor, For the task The corresponding normalized absolute value of the multidimensional compliance deviation, This represents the real-time backlog of tasks waiting to be processed at this node. The sensitivity of dynamic response to the business execution corresponding to this node.

9. The clinical trial end-to-end quality closed-loop control system according to claim 1, characterized in that, The process atomization control module divides clinical trials into subject screening, dosing follow-up, and endpoint assessment stages, and sets initial tolerances for compliance benchmarks based on the quality control accuracy requirements of each stage.

10. The clinical trial end-to-end quality closed-loop control system according to claim 1, characterized in that, The system also includes a global evidence link solidification unit, which encapsulates feedback gain adjustment instructions, data authenticity verification results, and the execution status of multi-level process control nodes into traceability metadata, and writes the traceability metadata into the distributed ledger in timestamp order.

Images