A clinical trial data intelligent acquisition method and system

Abstract

The application belongs to the field of data collection, and discloses a clinical test data intelligent collection method and system. The method comprises the following steps: acquiring monitoring data of multiple data sources and aligning the monitoring data to the same time axis to obtain multi-source aligned data; calculating a joint risk score based on the multi-source aligned data; according to the corresponding relationship between the joint risk score and multiple preset collection frequency gears, reducing the data collection frequency when the target gear is lower than the current gear; after the data collection frequency is reduced, sampling the multi-source aligned data during the reduction period and comparing the sampled data with the data before the reduction, calculating a quality difference value, and restoring the collection frequency to the gear before the reduction when the quality difference value exceeds a preset quality threshold. The system comprises an alignment module, a calculation module and a determination module. The application samples, compares and verifies the data collected during the frequency reduction after the data collection frequency is reduced, and automatically restores the collection frequency when the quality is degraded, thereby solving the problem that the prior art cannot guarantee the quality of the collected data when reducing the data collection frequency, and achieving the technical effect of reducing the frequency without reducing the quality.

Description

Technical Field

[0001] This invention relates to the field of data acquisition, and in particular to an intelligent method and system for acquiring clinical trial data. Background Technology

[0002] Clinical trial data collection is a fundamental step in ensuring the integrity and reliability of data during clinical trials. Traditional clinical trial data collection typically employs a fixed-frequency approach, periodically collecting monitoring data from subjects according to pre-set time cycles or visit nodes and entering the collected data into an electronic data acquisition system. This fixed-frequency approach cannot adapt to the varying risks across different stages and centers of a clinical trial, often resulting in high-input, low-efficiency collection activities. Therefore, a risk-based clinical trial data collection method has been proposed. This method dynamically adjusts the data collection frequency based on the risk status of the clinical trial: for example, assessing the risk level of each center based on historical incident statistics and developing a data collection and access plan accordingly; or, using a Bayesian model to assess the uncertainty of real-time collected data and increasing the data sampling frequency when the uncertainty exceeds a preset threshold. These methods, to a certain extent, achieve a linkage between data collection frequency and the risk status of the clinical trial.

[0003] However, existing risk-based clinical trial data collection methods still have shortcomings: when adjusting data collection frequency, current methods only focus on increasing the frequency when the risk level rises, i.e., increasing the data collection frequency when the risk assessment result is higher; however, they lack a mechanism to ensure the quality of data collected after the frequency reduction when the risk assessment result is lowered. When the data collection frequency is reduced, if the actual collected data shows quality deterioration such as an increase in the rate of missing key data items or an increase in the data error rate, existing methods cannot identify and respond to this quality deterioration in a timely manner, resulting in data collected after the frequency reduction failing to meet the data quality requirements for clinical trial statistical analysis. Therefore, how to ensure the quality of collected data while reducing the data collection frequency is a technical problem that existing technologies need to solve. Summary of the Invention

[0004] The purpose of this invention is to disclose an intelligent method and system for acquiring clinical trial data, thereby solving the technical problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: On the one hand, the present invention provides an intelligent method for acquiring clinical trial data, comprising: S1. Acquire monitoring data from multiple data sources in the clinical trial, and align the monitoring data from multiple data sources to the same timeline according to the timestamps carried by each monitoring data to obtain multi-source aligned data; S2, Calculate a joint risk score based on the multi-source aligned data, wherein the joint risk score is a numerical value used to characterize the overall risk level of multiple data sources; S3. Based on the correspondence between the joint risk score and multiple preset collection frequency levels, determine the target collection frequency level. When the target collection frequency level is lower than the current collection frequency level, lower the data collection frequency to the target collection frequency level. S4. After the data acquisition frequency is reduced, the first sample is extracted from the multi-source aligned data obtained during the reduction period according to the preset sampling rules. The first sample is compared with the corresponding part of the multi-source aligned data in the same period before the reduction, and the quality difference value is calculated. When the quality difference value exceeds the preset quality threshold, the data acquisition frequency is restored to the acquisition frequency level before the reduction.

[0006] Optionally, the step of aligning monitoring data from multiple data sources to the same timeline according to the timestamps carried by each monitoring data to obtain multi-source aligned data includes: S11, the time axis is divided into multiple time windows according to a preset time step, wherein the preset time step is greater than or equal to the minimum sampling interval of each data source; S12, The monitoring data from each data source are assigned to the time window containing the timestamp according to the timestamp they carry; S13, calculate the average value of multiple monitoring data from the same data source within the same time window, and determine the average value as the representative value of the data source within the time window; S14. For each data source within each time window, determine whether the data source has monitoring data within the time window; if not, determine that the representative value of the data source is missing within the time window, and fill the missing representative value by using the average value of the representative value of the data source in each time window within the preset historical statistical period. S15, combine the representative values ​​of each data source within each time window into the multi-source aligned data.

[0007] Optionally, calculating the joint risk score based on the multi-source aligned data includes: S21, for each representative value in the multi-source aligned data, determine whether it is within the corresponding preset normal range, wherein the preset normal range is the clinical normal value range of the data source to which the representative value belongs, and mark the representative value that exceeds the preset normal range as an abnormal value. S22, Calculate the ratio of the number of outliers to the total number of representative values ​​in the multi-source aligned data, and determine the ratio as the joint risk score.

[0008] Optionally, determining the target acquisition frequency level based on the correspondence between the joint risk score and multiple preset acquisition frequency levels includes: S31, multiple risk intervals are preset, wherein each risk interval is divided by a preset risk threshold, and each risk interval corresponds one-to-one with each collection frequency level; S32, determine the risk range in which the joint risk score is located; S33, the target acquisition frequency level is determined to be the acquisition frequency level corresponding to the risk interval where the joint risk score is located.

[0009] Optionally, the step of extracting the first sample from the multi-source aligned data obtained during the downsampling period according to a preset sampling rule includes: S41, Determine the preset sampling ratio; S42, Multiply the total number of representative values ​​in the multi-source aligned data obtained during the downsampling period by the preset sampling ratio to obtain the sampling quantity; S43, from the multi-source aligned data obtained during the downsizing period, representative values ​​are randomly selected according to the sampling quantity, and the selected representative values ​​are used to form the first sample.

[0010] Optionally, the step of comparing the first sample with the corresponding portion of the multi-source aligned data from the same time period before downsampling and calculating the quality difference value includes: S51, for each representative value in the first sample, a corresponding representative value is determined from the multi-source aligned data of the same period before the downsampling, wherein the corresponding representative value has the same time window order and the same data source as the representative value in the first sample. S52, perform a consistency comparison between each representative value in the first sample and the corresponding representative value; S53, count the number of inconsistent representative values, and determine the quality difference value as the ratio of the number of inconsistent representative values ​​to the total number of representative values ​​in the first sample.

[0011] Optionally, S3 further includes: S61, when the target acquisition frequency level is higher than the current acquisition frequency level, determine whether the joint risk score exceeds the preset upward adjustment threshold corresponding to the target acquisition frequency level, and whether this state continues for a first preset duration, wherein the preset upward adjustment threshold is greater than the upper limit of the risk range corresponding to the target acquisition frequency level; S62, if all the judgment results are yes, the data acquisition frequency is increased to the target acquisition frequency level.

[0012] Optionally, calculating the joint risk score based on the multi-source aligned data includes: S71, for each data source, determine whether each representative value of the data source is within the corresponding preset normal range, wherein the preset normal range is the clinical normal value interval corresponding to the data source, and mark the representative values ​​that exceed the preset normal range as abnormal values; S72, For each data source, calculate the ratio of the number of outliers in that data source to the total number of representative values ​​in that data source to obtain the sub-risk score for that data source; S73, obtain the preset weight coefficients of each data source, wherein the weight coefficients are positively correlated with the criticality of the data item corresponding to the data source in the clinical trial, and the sum of the weight coefficients of each data source is 1. S74, Multiply the sub-risk scores of each data source by their corresponding weight coefficients and sum them up, and determine the summation result as the joint risk score.

[0013] Optionally, after S4, it also includes: S81, Generate acquisition frequency adjustment warning information, wherein the acquisition frequency adjustment warning information includes the downgraded acquisition frequency level, the restored acquisition frequency level, and the quality difference value. S82, the acquisition frequency adjustment warning information is sent to the preset acquisition management terminal; S83, write the recordings of frequency reduction, frequency recovery, quality difference value, and the sending record of the frequency adjustment warning information into the audit log.

[0014] On the other hand, the present invention provides an intelligent clinical trial data acquisition system, comprising: The alignment module is used to acquire monitoring data from multiple data sources in clinical trials and align the monitoring data from multiple data sources to the same timeline according to the timestamps carried by each monitoring data to obtain multi-source aligned data. The calculation module is used to calculate a joint risk score based on the multi-source aligned data, wherein the joint risk score is a numerical value used to characterize the overall risk level of multiple data sources; The determination module is used to determine the target collection frequency level based on the correspondence between the joint risk score and multiple preset collection frequency levels. When the target collection frequency level is lower than the current collection frequency level, the data collection frequency is lowered to the target collection frequency level. The calculation module is also used to extract a first sample from the multi-source aligned data obtained during the downsampling period according to a preset sampling rule after the data acquisition frequency is reduced, and compare the first sample with the corresponding part of the multi-source aligned data in the same period before the downsampling to calculate the quality difference value. When the quality difference value exceeds a preset quality threshold, the data acquisition frequency is restored to the acquisition frequency level before the downsampling. Beneficial effects

[0015] This application solves the technical problem that existing technologies cannot guarantee the quality of collected data when reducing the data collection frequency by synergistically combining multi-source data time alignment, joint risk scoring, adaptive adjustment of collection frequency, and quality verification closed loop after frequency reduction.

[0016] Specifically, by aligning monitoring data from multiple data sources to the same timeline according to timestamps to obtain multi-source aligned data, the inconsistency in data granularity caused by asynchronous sampling times between different data sources is eliminated, providing a unified data foundation for subsequent joint risk scoring. Calculating the joint risk score based on multi-source aligned data comprehensively reflects the overall risk level of multiple data sources, making the assessment of clinical trial data collection risk no longer dependent on a single data source, thus improving the comprehensiveness and accuracy of risk assessment. The target collection frequency level is determined based on the correspondence between the joint risk score and multiple preset collection frequency levels. When the target level is lower than the current level, the data collection frequency is reduced, allowing the data collection frequency to adaptively decrease as the risk level decreases. This reduces the amount of data collected during the lower-risk phase of the clinical trial, lowering collection costs and the burden on subjects. After the data acquisition frequency is reduced, the quality difference value is calculated by sampling from the multi-source aligned data obtained during the reduction period and comparing it with the data from the same period before the reduction. This allows for a quantitative assessment of the actual impact of reducing the acquisition frequency on data quality. When the quality difference value exceeds the preset quality threshold, the acquisition frequency is automatically restored to the level before the reduction. This enables timely identification and automatic correction of data quality degradation caused by the frequency reduction, achieving the technical effect of continuously ensuring the quality of acquired data while reducing the data acquisition frequency.

[0017] Furthermore, by averaging monitoring data from the same data source within the same time window and filling in missing data according to preset rules, the impact of random fluctuations and missing data in the original monitoring data on subsequent scoring can be eliminated. By assigning weight coefficients to each data source and determining the joint risk score through weighted summation, the role of key data sources in risk assessment can be highlighted. By establishing a hysteresis interval mechanism that separates upward and downward adjustment thresholds and combines it with duration judgment, frequent fluctuations in the collection frequency can be avoided when the risk score is near the threshold, thus improving the stability of collection frequency adjustment. By writing collection frequency adjustment records, quality difference value comparison records, and early warning information sending records into the audit log, the traceability requirements of clinical trials for the data collection process can be met. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of an intelligent clinical trial data acquisition method according to the present invention.

[0020] Figure 2 This is a schematic diagram illustrating the calculation process of the joint risk score of the present invention.

[0021] Figure 3 This is a schematic diagram illustrating the process of determining the target acquisition frequency level according to the present invention.

[0022] Figure 4 This is a schematic diagram of an intelligent clinical trial data acquisition system according to the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, the present invention provides an intelligent method for acquiring clinical trial data, comprising: S1: Acquire monitoring data from multiple data sources in the clinical trial, and align the monitoring data from multiple data sources to the same timeline according to the timestamps carried by each monitoring data to obtain multi-source aligned data.

[0025] The data source is the system or device that generates monitoring data during the clinical trial. For example, the data source includes at least one of an electronic data acquisition system, a patient-reported outcome (PDA) terminal, a wearable monitoring device, and a laboratory testing system. The electronic data acquisition system records vital signs, physical examination data, and adverse event data of subjects at visit points; the PDA terminal receives symptom scores and medication adherence records self-reported by subjects; the wearable monitoring device continuously records physiological parameters such as heart rate, blood oxygen saturation, and steps; and the laboratory testing system outputs test data such as complete blood counts and biochemical indicators for subjects.

[0026] The monitoring data consists of individual data records generated by each data source. Each monitoring data record carries a timestamp, which is used to identify the sampling time of the monitoring data. For example, a monitoring data record in the electronic data acquisition system is [Subject ID, Visit Node, Systolic Blood Pressure, 130, Unit: mmHg, Timestamp: 2024-01-15 09:30:00]; a monitoring data record in the wearable monitoring device is [Subject ID, Heart Rate, 78, Unit: bpm, Timestamp: 2024-01-15 09:30:05]; a monitoring data record in the laboratory testing system is [Subject ID, White Blood Cell Count, 6.5, Unit: 10^9 / L, Timestamp: 2024-01-15 08:00:00].

[0027] The time axis is a unified time reference used to align monitoring data from various data sources, and the time axis uses a preset time step as its basic unit.

[0028] Optionally, monitoring data from multiple data sources in the clinical trial can be acquired, including: acquiring the monitoring data through preset data interfaces corresponding to each data source; and temporarily storing the acquired monitoring data according to data source and sampling time. For example, an electronic data acquisition system acquires monitoring data through a RESTful API interface, a patient-reported outcome acquisition terminal receives monitoring data filled in by subjects through a wireless network, wearable monitoring devices synchronize continuous recording of physiological parameters via Bluetooth or WiFi, and a laboratory testing system acquires test data through an HL7 interface; the acquired monitoring data is then sorted by data source identifier and timestamp and temporarily stored in a data buffer.

[0029] Optionally, the step of aligning monitoring data from multiple data sources to the same timeline according to the timestamps carried by each monitoring data to obtain multi-source aligned data includes: S11, the time axis is divided into multiple time windows according to a preset time step, wherein the preset time step is greater than or equal to the minimum sampling interval of each data source.

[0030] The preset time step is a fixed duration used to divide time windows. The value of the preset time step must be greater than or equal to the minimum sampling interval of each data source to ensure that each data source has at least one monitoring data point within each time window. For example, when the sampling interval of wearable monitoring devices is the smallest (5 minutes) and the sampling interval of the laboratory testing system is the largest (24 hours), the preset time step can be 24 hours. In this case, each data source will have at least one monitoring data point within each 24-hour time window, avoiding systematic data loss in certain data sources within some time windows.

[0031] The time window is a continuous period of time obtained by dividing the time axis into equal parts according to the preset time step. Adjacent time windows are connected end to end and do not overlap, and the length of the period covered by each time window is equal to the preset time step.

[0032] The minimum sampling interval is the shortest time interval between two consecutive samples from each data source in a clinical trial. For example, the minimum sampling interval for electronic data acquisition systems is 7 days (weekly visits), for patient-reported outcome acquisition terminals it is 1 day (daily reporting), for wearable monitoring devices it is 5 minutes (continuous heart rate monitoring), and for laboratory testing systems it is 28 days (blood samples every four weeks). The minimum sampling interval for each data source can be obtained from the clinical trial protocol.

[0033] Optionally, dividing the time axis into multiple time windows according to a preset time step includes: obtaining the minimum sampling interval of each data source; determining the maximum value from each minimum sampling interval and using the maximum value as the preset time step; and dividing the time axis from the start time of the clinical trial to the current time into multiple time windows according to the preset time step. For example, obtaining the minimum sampling interval of 5 minutes for wearable monitoring devices, 1 day for patient report outcome collection terminals, 7 days for electronic data acquisition systems, and 28 days for laboratory testing systems, where the maximum value is 28 days; using 28 days as the preset time step, starting from the start date of the clinical trial, 2024-01-01, dividing the time axis into multiple continuous time windows such as [2024-01-01, 2024-01-29), [2024-01-29, 2024-02-26), [2024-02-26, 2024-03-25), etc.

[0034] S12, the monitoring data from each data source are assigned to the time window containing the timestamp according to the timestamp they carry.

[0035] Optionally, the monitoring data from each data source is categorized into the time window containing the timestamp carried by each data source, including: Iterate through the monitoring data from each data source temporarily stored in the data buffer; For each piece of monitoring data, extract the timestamp it carries; The timestamp is compared with the start and end times of each time window to determine the time window in which the timestamp falls. The monitoring data is then incorporated into the defined time window.

[0036] For example, if a monitoring data point from a wearable monitoring device carries a timestamp of 2024-01-15 09:30:05, and this timestamp falls within the time window [2024-01-01, 2024-01-29), then the monitoring data will be included in that time window. Monitoring data from multiple data sources, including visit data from the electronic data acquisition system, symptom scores from the patient report outcome acquisition terminal, heart rate records from the wearable monitoring device, and blood routine data from the laboratory testing system, will also be included within the same time window.

[0037] S13: Calculate the average value of multiple monitoring data from the same data source within the same time window, and determine the average value as the representative value of the data source within the time window.

[0038] The representative value is the arithmetic mean of the monitoring data of the data source within the time window, which is used to replace multiple original monitoring data of the data source within the time window to eliminate random fluctuations in a single sampling.

[0039] Optionally, the step of averaging multiple monitoring data from the same data source within the same time window includes: For each data source within each time window, extract the numerical field of that data source from the monitoring data included in that time window; Calculate the arithmetic mean of the numeric fields in this data source; The arithmetic mean is determined as the representative value of the data source within the time window.

[0040] For example, within the time window [2024-01-01, 2024-01-29), the wearable monitoring device collected 4032 heart rate monitoring data points. The heart rate values ​​(unit: bpm) from each data point were extracted, and the arithmetic mean was calculated to be 75.3 bpm. Therefore, the representative value of this data source within the time window [2024-01-01, 2024-01-29) is determined to be 75.3 bpm. Similarly, if the laboratory testing system collected only one complete blood count data point within the same time window, then the white blood cell count value of 6.5 x 10^9 / L in that data point is the representative value of that data source within that time window.

[0041] S14. For each data source within each time window, determine whether the data source has monitoring data within the time window; if not, determine that the representative value of the data source is missing within the time window, and fill the missing representative value by using the average value of the representative values ​​of the data source in each time window within the preset historical statistical period.

[0042] The preset historical statistical period is a time period consisting of multiple consecutive time windows looking backward from the current time window. For example, the preset historical statistical period is the three time windows preceding the current time window.

[0043] Optionally, the step of filling missing representative values ​​by using the average of representative values ​​of each time window within a preset historical statistical period of the data source includes: determining the set of time windows covered by the preset historical statistical period; extracting representative values ​​of the data source within each time window from the set of time windows; calculating the arithmetic mean of the extracted representative values; and filling the arithmetic mean with the representative value of the data source within the current time window. For example, with a time step of 28 days and a preset historical statistical period of the past three time windows, the current time window is [2024-04-22, 2024-05-20), within which the laboratory testing system has no monitoring data; the representative values ​​of the laboratory testing system in the previous three time windows [2024-01-29, 2024-02-26), [2024-02-26, 2024-03-25), and [2024-03-25, 2024-04-22) are extracted as 6.5x10^9 / L, 6.8x10^9 / L, and 7.0x10^9 / L, respectively, and their arithmetic mean is calculated to be 6.77x10^9 / L. This value is then used as the representative value of the laboratory testing system in the current time window.

[0044] S15, combine the representative values ​​of each data source within each time window into the multi-source aligned data.

[0045] The multi-source aligned data is a two-dimensional data table with time windows as rows, data sources as columns, and representative values ​​as elements. Each row in the two-dimensional data table corresponds to a time window, each column corresponds to a data source, and each cell records the representative value of the data source within that time window.

[0046] Optionally, combining the representative values ​​of each data source within each time window into the multi-source aligned data includes: Construct a two-dimensional data table, where the number of rows in the two-dimensional data table is equal to the total number of time windows, and the number of columns is equal to the total number of data sources; For each time window and each data source, the representative value of the data source within that time window is filled into the corresponding row and column cells of the two-dimensional data table.

[0047] For example, the representative values ​​from four data sources (electronic data acquisition system, patient report outcome acquisition terminal, wearable monitoring device, and laboratory testing system) within three time windows are combined into the following multi-source aligned data: The first row corresponds to the time window [2024-01-01, 2024-01-29), with four cells recording representative values ​​(systolic blood pressure 128 mmHg, symptom score 3, heart rate 75.3 bpm, white blood cell count 6.5 x 10^9 / L); the second row corresponds to the time window [ The first row (2024-01-29, 2024-02-26) contains four cells recording representative values ​​(systolic blood pressure 132 mmHg, symptom score 2, heart rate 76.1 bpm, white blood cell count 6.8 x 10^9 / L); the second row (corresponding to the time window [2024-02-26, 2024-03-25)) contains four cells recording representative values ​​(systolic blood pressure 130 mmHg, symptom score 4, heart rate 74.8 bpm, white blood cell count 7.0 x 10^9 / L). The representative values ​​filled by S14 are not distinguished from the unfilled representative values ​​in the multi-source aligned data to ensure consistent processing in subsequent joint risk score calculations.

[0048] Compared with existing technologies, this embodiment aligns monitoring data from multiple data sources, such as electronic data acquisition systems, patient report outcome acquisition terminals, wearable monitoring devices, and laboratory testing systems, to the same timeline. During the alignment process, representative values ​​and filling rules are used to eliminate the asynchronous and random fluctuations in the sampling time of the original monitoring data. This allows the subsequent joint risk score to be built on a unified data granularity, avoiding the one-sidedness and instability caused by relying on a single data source or a single sampling value for risk assessment in existing technologies.

[0049] S2, Calculate a joint risk score based on the multi-source aligned data, wherein the joint risk score is a numerical value used to characterize the overall risk level of multiple data sources.

[0050] like Figure 2 As shown, the calculation of the joint risk score based on the multi-source aligned data includes: S21, for each representative value in the multi-source aligned data, determine whether it is within the corresponding preset normal range, wherein the preset normal range is the clinical normal value interval of the data source to which the representative value belongs, and mark the representative values ​​that exceed the preset normal range as abnormal values.

[0051] The preset normal range refers to the clinical normal value range corresponding to the data source. For example, the preset normal range for systolic blood pressure in an electronic data acquisition system is 90 to 140 mmHg, the preset normal range for symptom scores in a patient report outcome acquisition terminal is 0 to 3, the preset normal range for heart rate in a wearable monitoring device is 60 to 100 bpm, and the preset normal range for white blood cell count in a laboratory testing system is 4.0 to 10.0 x 10^9 / L.

[0052] Abnormal values ​​are representative values ​​that exceed the preset normal range. When a representative value is greater than the upper limit of the preset normal range or less than the lower limit of the preset normal range, the representative value is marked as an abnormal value; when a representative value falls within the preset normal range, the representative value is marked as a normal value.

[0053] Optionally, determining whether the value is within the corresponding preset normal range includes: extracting representative values ​​one by one from the multi-source aligned data; determining the data source to which the representative value belongs; querying the preset normal range corresponding to the data source from a pre-stored normal range table; and comparing the value of the representative value with the upper and lower limits of the preset normal range. For example, the multi-source aligned data records a representative value of 75.3 bpm for a wearable monitoring device within the time window [2024-01-01, 2024-01-29), and the preset normal range table records a heart rate of 60 to 100 bpm. 75.3 bpm falls within this range, so the representative value is marked as normal. Another example is a representative value of 132 mmHg systolic blood pressure from an electronic data acquisition system within the time window [2024-01-29, 2024-02-26]. g. The preset normal range for systolic blood pressure is 90 to 140 mmHg. 132 mmHg falls within this range, so this representative value is marked as normal. For example, in the laboratory testing system, the representative value for white blood cell count within the time window [2024-02-26, 2024-03-25] is 11.2 x 10^9 / L. The preset normal range for white blood cell count is 4.0 to 10.0 x 10^9 / L. 11.2 x 10^9 / L is greater than the upper limit of 10.0, so this representative value is marked as abnormal.

[0054] S22, Calculate the ratio of the number of outliers to the total number of representative values ​​in the multi-source aligned data, and determine the ratio as the joint risk score.

[0055] The joint risk score is a value between 0 and 1. The closer the joint risk score is to 1, the higher the overall risk level of multiple data sources.

[0056] Optionally, the ratio of the number of outliers to the total number of representative values ​​in the multi-source aligned data includes: traversing all representative values ​​in the multi-source aligned data; counting the number of representative values ​​marked as outliers; counting the total number of representative values ​​in the multi-source aligned data; and calculating the ratio of the number of outliers to the total number of representative values. For example, in multi-source aligned data from 3 time windows and 4 data sources, there are a total of 12 representative values. Among them, S21 marks 1 outlier (white blood cell count 11.2 x 10^9 / L) and 11 normal values. Then the number of outliers is 1, the total number of representative values ​​is 12, and the joint risk score = 1 / 12 = 0.083. This score is close to 0, indicating that the overall risk level of the current multi-source data sources is low, and the data collection frequency can be appropriately reduced.

[0057] Optionally, calculating the joint risk score based on the multi-source aligned data includes: S71, for each data source, determine whether each representative value of the data source is within the corresponding preset normal range, wherein the preset normal range is the clinical normal value interval corresponding to the data source, and mark the representative values ​​that exceed the preset normal range as abnormal values.

[0058] Optionally, determining whether each representative value of the data source is within the corresponding preset normal range includes: extracting all representative values ​​of each data source column by column from the multi-source aligned data; For each data source, all representative values ​​are compared one by one with the preset normal range corresponding to that data source.

[0059] For example, for the data source of wearable monitoring devices, representative values ​​of 75.3 bpm, 76.1 bpm, and 74.8 bpm within three time windows were extracted and compared with the preset normal range of heart rate of 60 to 100 bpm. All three representative values ​​fell within the normal range and were marked as normal values. For the data source of the laboratory testing system, representative values ​​of 6.5 x 10^9 / L, 6.8 x 10^9 / L, and 11.2 x 10^9 / L were extracted. Among them, 11.2 x 10^9 / L exceeded the preset normal range of white blood cell count of 4.0 to 10.0 x 10^9 / L and was marked as an abnormal value. The rest were marked as normal values.

[0060] S72, For each data source, calculate the ratio of the number of outliers in that data source to the total number of representative values ​​in that data source to obtain the sub-risk score for that data source.

[0061] The sub-risk score is a value between 0 and 1. A higher sub-risk score indicates a higher proportion of anomalies in the data source.

[0062] Optionally, the step of calculating the ratio of the number of outliers in the data source to the total number of representative values ​​in the data source includes: for each data source, counting the number of representative values ​​marked as outliers in the data source; counting the total number of representative values ​​in the data source across all time windows in the multi-source aligned data; and calculating the ratio of the number of outliers to the total number of representative values ​​as a sub-risk score for the data source. For example, the electronic data acquisition system has three representative values ​​(systolic blood pressure 128, 132, 130 mmHg) within three time windows, all of which are within the preset normal range, with zero outliers and a sub-risk score of 0 / 3 = 0; the patient report outcome acquisition terminal has three representative values ​​(symptom scores 3, 2, 4), with symptom score 4 exceeding the preset normal range of 0 to 3, resulting in one outlier and a sub-risk score of 1 / 3 = 0.33; the wearable monitoring device has three representative values ​​(heart rate 75.3, 76.1, 74.8 bpm), all of which are within the preset normal range, with zero outliers and a sub-risk score of 0 / 3 = 0; the laboratory testing system has three representative values ​​(white blood cell count 6.5, 6.8, 11.2 x 10^9 / L), with 11.2 exceeding the preset normal range, resulting in one outlier and a sub-risk score of 1 / 3 = 0.33.

[0063] S73, obtain the preset weight coefficients of each data source, wherein the weight coefficients are positively correlated with the criticality of the data item corresponding to the data source in the clinical trial, and the sum of the weight coefficients of each data source is 1.

[0064] The weighting coefficient represents the importance of the data source in the joint risk score. A larger weighting coefficient indicates a greater contribution of the sub-risk scores of that data source to the joint risk score. The weighting coefficient is positively correlated with the criticality of the data items from that data source in the clinical trial. For example, in clinical trials evaluating drug safety, the criticality of laboratory testing systems (complete blood counts, biochemical indicators) and electronic data acquisition systems (vital signs, adverse events) is higher than that of patient-reported outcome acquisition terminals and wearable monitoring devices. Therefore, the weighting coefficients for laboratory testing systems and electronic data acquisition systems can be set to larger values.

[0065] Optionally, obtaining the preset weight coefficients for each data source includes: Query the corresponding weight coefficient from the pre-stored weight configuration table according to the data source identifier.

[0066] For example, the pre-stored weight configuration table is as follows: the weight coefficient for the electronic data acquisition system is 0.30, the weight coefficient for the patient reported outcome acquisition terminal is 0.20, the weight coefficient for the wearable monitoring device is 0.20, and the weight coefficient for the laboratory testing system is 0.30. The sum of the four weight coefficients is 0.30 + 0.20 + 0.20 + 0.30 = 1. This weight configuration table can be pre-set by the trial designer through a configuration interface before the clinical trial begins, based on the criticality of the data items in the trial protocol.

[0067] S74, Multiply the sub-risk scores of each data source by their corresponding weight coefficients and sum them up, and determine the summation result as the joint risk score.

[0068] Optionally, the step of multiplying the sub-risk scores of each data source by their corresponding weighting coefficients and then summing them includes: for each data source, multiplying the sub-risk score of that data source by its weighting coefficient to obtain a weighted sub-risk score for that data source; and summing the weighted sub-risk scores of all data sources to obtain the joint risk score. For example, the sub-risk score of the electronic data acquisition system is 0 multiplied by a weighting coefficient of 0.30, which equals 0; the sub-risk score of the patient reported outcome acquisition terminal is 0.33 multiplied by a weighting coefficient of 0.20, which equals 0.066; the sub-risk score of the wearable monitoring device is 0 multiplied by a weighting coefficient of 0.20, which equals 0; the sub-risk score of the laboratory testing system is 0.33 multiplied by a weighting coefficient of 0.30, which equals 0.099; the joint risk score = 0 + 0.066 + 0 + 0.099 = 0.165.

[0069] Compared with existing single-indicator uncertainty assessment methods based on Bayesian models, the joint risk score in this embodiment integrates representative values ​​from various data sources. Through outlier labeling and optional data source weight coefficient configuration, it can simultaneously reflect the risk contribution of multiple data sources within the same scoring framework. This makes the judgment of the risk of clinical trial data collection no longer limited to fluctuations in a single dimension, thus improving the accuracy and comprehensiveness of risk assessment.

[0070] S3. Based on the correspondence between the joint risk score and the preset multiple collection frequency levels, determine the target collection frequency level. When the target collection frequency level is lower than the current collection frequency level, lower the data collection frequency to the target collection frequency level.

[0071] like Figure 3 As shown, determining the target acquisition frequency level based on the correspondence between the joint risk score and multiple preset acquisition frequency levels includes: S31, multiple risk intervals are preset, wherein each risk interval is divided by a preset risk threshold, and each risk interval corresponds one-to-one with each acquisition frequency level.

[0072] The risk range is a continuous numerical range separated by a preset risk threshold within the range of the joint risk score.

[0073] The preset risk threshold is the critical value used to divide the risk range. For example, the value range of the joint risk score is 0 to 1, and the preset risk threshold includes 0.15 and 0.35, thus dividing the risk range into 3 risk ranges: low risk range [0, 0.15), medium risk range [0.15, 0.35], and high risk range [0.35, 1].

[0074] The data collection frequency tiers represent a tiered configuration of data collection frequencies. There are three frequency tiers that correspond one-to-one with three risk ranges: low-risk range corresponds to low-frequency tiers, medium-risk range corresponds to medium-frequency tiers, and high-risk range corresponds to high-frequency tiers. For example, the low-frequency tier collects data every 48 hours, the medium-frequency tier collects data every 24 hours, and the high-frequency tier collects data every 4 hours.

[0075] Optionally, the preset multiple risk intervals include: obtaining the number n of preset risk thresholds; dividing the value range [0, 1] of the joint risk score into n+1 risk intervals according to the n preset risk thresholds; assigning a corresponding collection frequency level to each risk interval, establishing a one-to-one correspondence between the risk interval and the collection frequency level, and storing the correspondence in a level mapping table. For example, n=2 preset risk thresholds 0.15 and 0.35 divide the value range into 3 risk intervals, and the records stored in the level mapping table are: [low risk interval, low frequency level], [medium risk interval, medium frequency level], [high risk interval, high frequency level].

[0076] S32, determine the risk range in which the joint risk score is located.

[0077] Optionally, determining the risk interval in which the joint risk score falls includes: comparing the joint risk score with the upper and lower limits of each risk interval one by one; when the joint risk score is greater than or equal to the lower limit of a certain risk interval and less than the upper limit of that risk interval, it is determined that the joint risk score falls into that risk interval. For example, if the joint risk score calculated based on the S74 weighted scheme is 0.165, comparing 0.165 with the low-risk interval [0, 0.15), 0.165>=0.15, so it does not fall into the low-risk interval; comparing it with the medium-risk interval [0.15, 0.35), 0.165>=0.15 and 0.165<0.35, it is determined that the joint risk score 0.165 falls into the medium-risk interval.

[0078] S33, the target acquisition frequency level is determined to be the acquisition frequency level corresponding to the risk interval where the joint risk score is located.

[0079] The target data collection frequency level is the data collection frequency configuration corresponding to the current joint risk score.

[0080] Optionally, determining the target acquisition frequency level as the acquisition frequency level corresponding to the risk interval where the joint risk score is located includes: querying the acquisition frequency level corresponding to the risk interval where the joint risk score is located from the level mapping table; and determining the acquisition frequency level as the target acquisition frequency level. For example, if the joint risk score of 0.165 falls into the medium-risk interval, and the acquisition frequency level corresponding to the medium-risk interval is found to be the medium-frequency level from the level mapping table, then the target acquisition frequency level is determined to be the medium-frequency level.

[0081] Optionally, after determining the target acquisition frequency level, compare the target acquisition frequency level with the current acquisition frequency level. For example, if the current acquisition frequency level is a high-frequency level and the target acquisition frequency level is a medium-frequency level, and the target acquisition frequency level is lower than the current acquisition frequency level, the downgrading condition is met, and the data acquisition frequency is downgraded from the current high-frequency level (acquiring once every 4 hours) to the target medium-frequency level (acquiring once every 24 hours); if the current acquisition frequency level is a low-frequency level and the target acquisition frequency level is a medium-frequency level, and the target acquisition frequency level is higher than the current acquisition frequency level, then the upgrading judgment process of S61-S62 is triggered.

[0082] Compared to existing technologies that only increase data collection frequency when risk increases, this embodiment automatically reduces data collection frequency when the combined risk score decreases by pre-setting a one-to-one correspondence between risk ranges and collection frequency levels. This allows data collection activities to be dynamically reduced as the risk level decreases, thereby reducing unnecessary data collection during the low-risk phase of clinical trials, lowering collection costs and the burden on subjects.

[0083] Optionally, S3 further includes: S61, when the target acquisition frequency level is higher than the current acquisition frequency level, determine whether the joint risk score exceeds the preset upward adjustment threshold corresponding to the target acquisition frequency level, and whether this state continues for a first preset duration, wherein the preset upward adjustment threshold is greater than the upper limit of the risk range corresponding to the target acquisition frequency level.

[0084] The preset upward adjustment threshold is a critical value for the joint risk score used to trigger an increase in the collection frequency. The preset upward adjustment threshold is greater than the upper limit of the risk range corresponding to the target collection frequency level, creating a hysteresis interval between the upward adjustment threshold and the lower limit threshold (i.e., the lower limit of the risk range). This prevents frequent switching of the collection frequency when the joint risk score fluctuates near the boundary of the risk range. For example, the medium-risk range corresponding to the medium-frequency level is [0.15, 0.35), with a lower limit of 0.15 (used to trigger the judgment of downgrading from medium frequency to low frequency) and an upper limit of 0.35. The preset upward adjustment threshold corresponding to the medium-frequency level is set to 0.40, which is greater than the upper limit of 0.35.

[0085] The first preset duration is the minimum duration during which the joint risk score must continuously exceed a preset upward adjustment threshold before triggering an upward adjustment. The first preset duration is measured in the number of time windows. For example, if the first preset duration is set to 3 time windows, and the time step is 28 days, the first preset duration corresponds to 84 days.

[0086] Optionally, determining whether the joint risk score exceeds a preset upward adjustment threshold and whether this state persists for a first preset duration includes: calculating the joint risk score for the current time window; comparing the joint risk score with the preset upward adjustment threshold corresponding to the target acquisition frequency level; if the joint risk score exceeds the preset upward adjustment threshold, incrementing the count for the duration exceeding the threshold by 1; if the joint risk score does not exceed the preset upward adjustment threshold, resetting the count for the duration exceeding the threshold to zero; and determining whether the count for the duration exceeding the threshold reaches the number of time windows corresponding to the first preset duration. For example, the current acquisition frequency level is a medium frequency level, the target acquisition frequency level is a high frequency level, the risk range corresponding to the high frequency level is [0.35, 1], its upper limit is 1, and the preset upward adjustment threshold is set to 0.40. The combined risk scores for the four consecutive time windows are 0.38, 0.42, 0.45, and 0.43, respectively. In the first window (0.38 < 0.40), the continuous count is reset to zero. In the second window (0.42 > 0.40), the continuous count is recorded as 1. In the third window (0.45 > 0.40), the continuous count is recorded as 2. In the fourth window (0.43 > 0.40), the continuous count is recorded as 3. At this point, the continuous count has reached the first preset duration of three time windows, and the judgment result is positive for all of them.

[0087] S62, if all the judgment results are yes, the data acquisition frequency is increased to the target acquisition frequency level.

[0088] Optionally, increasing the data acquisition frequency to the target acquisition frequency level includes: when the continuous count exceeding the threshold reaches the number of time windows corresponding to the first preset duration, modifying the acquisition parameters of each data source, and adjusting the acquisition interval of each data source to the acquisition interval corresponding to the target acquisition frequency level. For example, if the current acquisition frequency level is the medium frequency level (each data source acquires data once every 24 hours), and the target acquisition frequency level is the high frequency level, the acquisition parameters of each data source are modified to acquire data once every 4 hours. After the increase is completed, the continuous count exceeding the threshold is cleared to prepare for the next increase judgment.

[0089] Compared with the frequency adjustment triggering method based on a single threshold in the prior art, this embodiment introduces a hysteresis interval in the upward adjustment direction by setting a preset upward adjustment threshold greater than the upper limit of the risk interval and combining it with the continuous judgment of the first preset duration. This effectively avoids the frequent switching of the collection frequency caused by short-term fluctuations of the joint risk score near the level boundary, and improves the stability of the collection frequency adjustment.

[0090] S4. After the data acquisition frequency is reduced, the first sample is extracted from the multi-source aligned data obtained during the reduction period according to the preset sampling rules. The first sample is compared with the corresponding part of the multi-source aligned data in the same period before the reduction, and the quality difference value is calculated. When the quality difference value exceeds the preset quality threshold, the data acquisition frequency is restored to the acquisition frequency level before the reduction.

[0091] Optionally, the step of extracting the first sample from the multi-source aligned data obtained during the downsampling period according to a preset sampling rule includes: S41, Determine the preset sampling ratio.

[0092] The preset sampling ratio is the percentage of the first sample drawn from the multi-source aligned data obtained during the downsampling period.

[0093] Optionally, determining the preset sampling ratio includes: querying the value of the preset sampling ratio from a pre-stored sampling configuration table. For example, the pre-stored sampling configuration table records a preset sampling ratio of 20%, which means that 20% of the representative values ​​are extracted from the multi-source aligned data obtained during the downsampling period as the first sample.

[0094] Optionally, the preset sampling ratio is related to the preset quality threshold. When the preset quality threshold is set low (low tolerance for quality differences), the preset sampling ratio is increased accordingly to ensure that the first sample has sufficient statistical representativeness. For example, when the preset quality threshold is 5%, the preset sampling ratio is set to 30%; when the preset quality threshold is 10%, the preset sampling ratio is set to 20%.

[0095] S42, multiply the total number of representative values ​​in the multi-source aligned data obtained during the downsampling period by the preset sampling ratio to obtain the sampling quantity.

[0096] Optionally, multiplying the total number of representative values ​​in the multi-source aligned data obtained during the downsizing period by a preset sampling ratio includes: counting the number of time windows during the downsizing period; obtaining the number of data sources; multiplying the number of time windows by the number of data sources to obtain the total number of representative values; multiplying the total number of representative values ​​by the preset sampling ratio to obtain the sample size; if the product is not an integer, rounding the product up. For example, if the downsizing period covers 3 time windows, the number of data sources is 4, the total number of representative values ​​= 3 x 4 = 12, the preset sampling ratio is 20%, the sample size = 12 x 20% = 2.4, and after rounding up, the sample size is 3, indicating that 3 representative values ​​are randomly selected from these 12 representative values ​​to form the first sample.

[0097] S43, from the multi-source aligned data obtained during the downsizing period, representative values ​​are randomly selected according to the sampling quantity, and the selected representative values ​​are used to form the first sample.

[0098] The first sample is a set of representative values ​​extracted from the multi-source aligned data during the downsampling period, used to evaluate the impact of downsampling on data quality by comparing it with data from the same period before downsampling.

[0099] Optionally, the step of randomly selecting representative values ​​according to the sampling quantity includes: arranging all representative values ​​in the multi-source aligned data obtained during the downsizing period into a list according to the time window and data source; using a random number generator to randomly select representative values ​​from the list without replacement, equal to the sampling quantity; forming a first sample from the selected representative values, and recording the time window and data source to which each selected representative value belongs. For example, during the downsizing period, there are 12 representative values ​​from 3 time windows and 4 data sources, with a sampling quantity of 3. The randomly selected representative values ​​are 74.8 (wearable monitoring device, time window [2024-02-26, 2024-03-25)), 132 (electronic data acquisition system, time window [2024-01-29, 2024-02-26)), and 4 (patient reported outcome acquisition terminal, time window [2024-02-26, 2024-03-25)), and these 3 representative values ​​form the first sample.

[0100] Optionally, the step of comparing the first sample with the corresponding portion of the multi-source aligned data from the same time period before downsampling and calculating the quality difference value includes: S51, for each representative value in the first sample, a corresponding representative value is determined from the multi-source aligned data of the same period before the downsampling, wherein the corresponding representative value has the same time window position and the same data source as the representative value in the first sample.

[0101] The corresponding representative value is the representative value in the multi-source aligned data before downscaling that has the same data source as the representative value in the first sample and corresponds to the same time window sequence. The time window sequence indicates the order number of the time window during the clinical trial. Time windows with the same sequence before and after downscaling cover the same time period length but correspond to different calendar periods. For example, the first time window after downscaling and the last time window before downscaling are both the first windows of the data source within their respective collection periods, and they have the same sequence, forming a corresponding relationship.

[0102] Optionally, determining the corresponding representative value includes: for each representative value in the first sample, obtaining the data source to which the representative value belongs and its time window position; from the multi-source aligned data before the downsizing, finding a representative value with the same data source and the same time window position as the representative value, and using it as the corresponding representative value. For example, in the first sample, there is a representative value of 74.8 bpm, which belongs to the wearable monitoring device data source and is located in the third time window after the downsizing; from the multi-source aligned data of the same period before the downsizing, finding the representative value of the wearable monitoring device data source in the third time window, we find 74.3 bpm, so 74.3 bpm is the corresponding representative value. As another example, in the first sample, there is a representative value 4 (symptom score), which belongs to the patient-reported outcome collection terminal data source and is located in the third time window after the downsizing; from the multi-source aligned data before the downsizing, finding the representative value of the patient-reported outcome collection terminal data source in the third time window, we find 2, so 2 is the corresponding representative value.

[0103] S52, perform a consistency comparison between each representative value in the first sample and the corresponding representative value.

[0104] Optionally, the consistency comparison includes: for each representative value in the first sample, comparing the value of the representative value with the value of the corresponding representative value. When the value of the representative value is equal to the value of the corresponding representative value, the representative value is determined to be consistent; when the value of the representative value is not equal to the value of the corresponding representative value, the representative value is determined to be inconsistent. For example, in the first sample, the representative value is 74.8 bpm (wearable monitoring device), and the corresponding representative value is 74.3 bpm; they are not equal, so they are determined to be inconsistent. Similarly, in the first sample, the representative value is 132 mmHg (electronic data acquisition system), and the corresponding representative value is 132 mmHg; they are equal, so they are determined to be consistent. In the first sample, the representative value is 4 (patient report outcome acquisition terminal), and the corresponding representative value is 2; they are not equal, so they are determined to be inconsistent.

[0105] S53, count the number of inconsistent representative values, and determine the quality difference value as the ratio of the number of inconsistent representative values ​​to the total number of representative values ​​in the first sample.

[0106] The quality difference value is between 0 and 1. The larger the quality difference value, the greater the deviation of the data collected after the sampling frequency is reduced compared with that before the reduction.

[0107] Optionally, the method of counting the number of inconsistent representative values ​​includes: traversing the S52 consistency comparison results and counting the number of representative values ​​judged as inconsistent; counting the total number of representative values ​​in the first sample; and calculating the ratio of the number of inconsistent representative values ​​to the total number of representative values. For example, if the first sample contains 3 representative values, in the S52 comparison, 74.8 bpm is judged as inconsistent, 132 mmHg is judged as consistent, and 4 is judged as inconsistent, the number of inconsistent representative values ​​is 2, the total number of representative values ​​is 3, and the quality difference value = 2 / 3 = 0.667. A quality difference value of 0.667 exceeds the preset quality threshold of 0.05 (i.e., 5%), indicating that the data quality has significantly deteriorated after the collection frequency was lowered, and the data collection frequency needs to be restored to the level before the reduction. Alternatively, if the quality difference value in another comparison is 0.03, which does not exceed the preset quality threshold, then the current reduced collection frequency level should be maintained.

[0108] Compared to existing technologies, this embodiment does not simply rely on the quality of the reduced data acquisition frequency after it is lowered. Instead, it extracts a first sample from the multi-source aligned data during the downsampling period according to a preset sampling rule. The representative value in this first sample is then compared pairwise with the representative values ​​in the same time window before the downsampling, and a quality difference value is calculated. Only when the quality difference value exceeds a preset quality threshold is the acquisition frequency restored to the level before the downsampling. This closed-loop quality verification mechanism after frequency reduction fills the gap in existing technologies that lack quality assurance measures during frequency reduction, ensuring that the reduction in data acquisition frequency does not compromise the quality of the acquired data, achieving the technical effect of cost reduction without compromising quality.

[0109] Optionally, after S4, it also includes: S81, Generate acquisition frequency adjustment warning information, wherein the acquisition frequency adjustment warning information includes the downgraded acquisition frequency level, the restored acquisition frequency level, and the quality difference value.

[0110] Among them, the data acquisition frequency adjustment early warning information is a record used to notify data acquisition managers that the data acquisition frequency will be automatically restored after the quality deteriorates due to frequency reduction.

[0111] Optionally, generating the acquisition frequency adjustment warning information includes: obtaining the reduced acquisition frequency level and the restored acquisition frequency level from the acquisition frequency adjustment record; obtaining the quality difference value from the quality difference value calculation result of S53; and combining the reduced acquisition frequency level, the restored acquisition frequency level, and the quality difference value into the acquisition frequency adjustment warning information. For example, if the reduced acquisition frequency level is a low-frequency level (acquiring once every 48 hours), the restored acquisition frequency level is a medium-frequency level (acquiring once every 24 hours), and the quality difference value is 0.667, the generated acquisition frequency adjustment warning information includes the above three data items.

[0112] S82, the acquisition frequency adjustment warning information is sent to the preset acquisition management terminal.

[0113] The data acquisition and management terminal is a receiving device used by clinical trial monitors or data managers to receive and display the early warning information generated by this method.

[0114] Optionally, sending the frequency adjustment warning information to a preset data acquisition management terminal includes: querying the communication address of the data acquisition management terminal from a pre-stored terminal address table; encapsulating the frequency adjustment warning information into a warning message according to a predetermined message format; and sending the warning message to the communication address through a preset communication channel. For example, if the communication address recorded in the terminal address table is the email address of the monitoring personnel, the warning message is encapsulated according to an email format and sent to that email address via SMTP. The content of the warning message is: "Frequency Acquisition Warning: On April 22, 2024, the frequency was restored from low frequency to mid frequency, with a quality difference value of 0.667. Please check the data acquisition situation during the frequency reduction period."

[0115] S83, write the recordings of frequency reduction, frequency recovery, quality difference value, and the sending record of the frequency adjustment warning information into the audit log.

[0116] The audit log is a time-series record file used to record key operations during the execution of this method. Each record in the audit log contains an operation timestamp and operation content.

[0117] Optionally, writing to the audit log includes: obtaining the current system time as the operation timestamp; generating a sampling frequency reduction record, wherein the sampling frequency reduction record includes the operation timestamp, the sampling frequency level before reduction, the sampling frequency level after reduction, and the value of the joint risk score at the time of reduction; generating a sampling frequency recovery record, wherein the sampling frequency recovery record includes the operation timestamp, the sampling frequency level before recovery, the sampling frequency level after recovery, and the quality difference value; generating an early warning sending record, wherein the early warning sending record includes the operation timestamp, the communication address of the receiving terminal, and the sending status; and appending the sampling frequency reduction record, the sampling frequency recovery record, and the early warning sending record to the audit log in timestamp order. For example, the three records in the audit log are as follows: "2024-03-25 00:00:00 Collection frequency downgraded to high frequency level → medium frequency level, joint risk score 0.165"; "2024-04-22 00:00:00 Collection frequency restored to medium frequency level → high frequency level, quality difference value 0.667"; "2024-04-22 00:00:05 Warning sent to the recipient's monitoring personnel email address, sent successfully".

[0118] like Figure 4 As shown, the present invention provides an intelligent clinical trial data acquisition system, comprising: The alignment module is used to acquire monitoring data from multiple data sources in clinical trials and align the monitoring data from multiple data sources to the same timeline according to the timestamps carried by each monitoring data to obtain multi-source aligned data. The calculation module is used to calculate a joint risk score based on the multi-source aligned data, wherein the joint risk score is a numerical value used to characterize the overall risk level of multiple data sources; The determination module is used to determine the target collection frequency level based on the correspondence between the joint risk score and multiple preset collection frequency levels. When the target collection frequency level is lower than the current collection frequency level, the data collection frequency is lowered to the target collection frequency level. The calculation module is also used to extract a first sample from the multi-source aligned data obtained during the downsampling period according to a preset sampling rule after the data acquisition frequency is reduced, and compare the first sample with the corresponding part of the multi-source aligned data in the same period before the downsampling to calculate the quality difference value. When the quality difference value exceeds a preset quality threshold, the data acquisition frequency is restored to the acquisition frequency level before the downsampling.

[0119] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for intelligent acquisition of clinical trial data, characterized in that, include: S1. Acquire monitoring data from multiple data sources in the clinical trial, and align the monitoring data from multiple data sources to the same timeline according to the timestamps carried by each monitoring data to obtain multi-source aligned data; S2, Calculate a joint risk score based on the multi-source aligned data, wherein the joint risk score is a numerical value used to characterize the overall risk level of multiple data sources; S3. Based on the correspondence between the joint risk score and multiple preset collection frequency levels, determine the target collection frequency level. When the target collection frequency level is lower than the current collection frequency level, lower the data collection frequency to the target collection frequency level. S4. After the data acquisition frequency is reduced, the first sample is extracted from the multi-source aligned data obtained during the reduction period according to the preset sampling rules. The first sample is compared with the corresponding part of the multi-source aligned data in the same period before the reduction, and the quality difference value is calculated. When the quality difference value exceeds the preset quality threshold, the data acquisition frequency is restored to the acquisition frequency level before the reduction.

2. The intelligent acquisition method for clinical trial data according to claim 1, characterized in that, The step of aligning monitoring data from multiple data sources to the same timeline according to the timestamps carried by each monitoring data point to obtain multi-source aligned data includes: S11, the time axis is divided into multiple time windows according to a preset time step, wherein the preset time step is greater than or equal to the minimum sampling interval of each data source; S12, The monitoring data from each data source are assigned to the time window containing the timestamp according to the timestamp they carry; S13, calculate the average value of multiple monitoring data from the same data source within the same time window, and determine the average value as the representative value of the data source within the time window; S14. For each data source within each time window, determine whether the data source has monitoring data within the time window; if not, determine that the representative value of the data source is missing within the time window, and fill the missing representative value by using the average value of the representative value of the data source in each time window within the preset historical statistical period. S15, combine the representative values ​​of each data source within each time window into the multi-source aligned data.

3. The intelligent acquisition method for clinical trial data according to claim 1, characterized in that, The calculation of the joint risk score based on the multi-source aligned data includes: S21, for each representative value in the multi-source aligned data, determine whether it is within the corresponding preset normal range, wherein the preset normal range is the clinical normal value range of the data source to which the representative value belongs, and mark the representative value that exceeds the preset normal range as an abnormal value. S22, Calculate the ratio of the number of outliers to the total number of representative values ​​in the multi-source aligned data, and determine the ratio as the joint risk score.

4. The intelligent acquisition method for clinical trial data according to claim 1, characterized in that, The step of determining the target acquisition frequency level based on the correspondence between the joint risk score and multiple preset acquisition frequency levels includes: S31, multiple risk intervals are preset, wherein each risk interval is divided by a preset risk threshold, and each risk interval corresponds one-to-one with each collection frequency level; S32, determine the risk range in which the joint risk score is located; S33, the target acquisition frequency level is determined to be the acquisition frequency level corresponding to the risk interval where the joint risk score is located.

5. The intelligent acquisition method for clinical trial data according to claim 2, characterized in that, The step of extracting the first sample from the multi-source aligned data obtained during the downsampling period according to a preset sampling rule includes: S41, Determine the preset sampling ratio; S42, Multiply the total number of representative values ​​in the multi-source aligned data obtained during the downsampling period by the preset sampling ratio to obtain the sampling quantity; S43, from the multi-source aligned data obtained during the downsizing period, representative values ​​are randomly selected according to the sampling quantity, and the selected representative values ​​are used to form the first sample.

6. The intelligent acquisition method for clinical trial data according to claim 2, characterized in that, The step of comparing the first sample with the corresponding portion of the multi-source aligned data from the same time period before downsampling and calculating the quality difference value includes: S51, for each representative value in the first sample, a corresponding representative value is determined from the multi-source aligned data of the same period before the downsampling, wherein the corresponding representative value has the same time window order and the same data source as the representative value in the first sample. S52, perform a consistency comparison between each representative value in the first sample and the corresponding representative value; S53, count the number of inconsistent representative values, and determine the quality difference value as the ratio of the number of inconsistent representative values ​​to the total number of representative values ​​in the first sample.

7. The intelligent acquisition method for clinical trial data according to claim 4, characterized in that, The S3 further includes: S61, when the target acquisition frequency level is higher than the current acquisition frequency level, determine whether the joint risk score exceeds the preset upward adjustment threshold corresponding to the target acquisition frequency level, and whether this state continues for a first preset duration, wherein the preset upward adjustment threshold is greater than the upper limit of the risk range corresponding to the target acquisition frequency level; S62, if all the judgment results are yes, the data acquisition frequency is increased to the target acquisition frequency level.

8. The intelligent acquisition method for clinical trial data according to claim 1, characterized in that, The calculation of the joint risk score based on the multi-source aligned data includes: S71, for each data source, determine whether each representative value of the data source is within the corresponding preset normal range, wherein the preset normal range is the clinical normal value interval corresponding to the data source, and mark the representative values ​​that exceed the preset normal range as abnormal values; S72, For each data source, calculate the ratio of the number of outliers in that data source to the total number of representative values ​​in that data source to obtain the sub-risk score for that data source; S73, obtain the preset weight coefficients of each data source, wherein the weight coefficients are positively correlated with the criticality of the data item corresponding to the data source in the clinical trial, and the sum of the weight coefficients of each data source is 1. S74, Multiply the sub-risk scores of each data source by their corresponding weight coefficients and sum them up, and determine the summation result as the joint risk score.

9. The intelligent acquisition method for clinical trial data according to claim 1, characterized in that, Following S4, it also includes: S81, Generate acquisition frequency adjustment warning information, wherein the acquisition frequency adjustment warning information includes the downgraded acquisition frequency level, the restored acquisition frequency level, and the quality difference value. S82, the acquisition frequency adjustment warning information is sent to the preset acquisition management terminal; S83, write the recordings of frequency reduction, frequency recovery, quality difference value, and the sending record of the frequency adjustment warning information into the audit log.

10. A clinical trial data intelligent acquisition system, characterized in that, include: The alignment module is used to acquire monitoring data from multiple data sources in clinical trials and align the monitoring data from multiple data sources to the same timeline according to the timestamps carried by each monitoring data to obtain multi-source aligned data. The calculation module is used to calculate a joint risk score based on the multi-source aligned data, wherein the joint risk score is a numerical value used to characterize the overall risk level of multiple data sources; The determination module is used to determine the target collection frequency level based on the correspondence between the joint risk score and multiple preset collection frequency levels. When the target collection frequency level is lower than the current collection frequency level, the data collection frequency is lowered to the target collection frequency level. The calculation module is also used to extract a first sample from the multi-source aligned data obtained during the downsampling period according to a preset sampling rule after the data acquisition frequency is reduced, and compare the first sample with the corresponding part of the multi-source aligned data in the same period before the downsampling to calculate the quality difference value. When the quality difference value exceeds a preset quality threshold, the data acquisition frequency is restored to the acquisition frequency level before the downsampling.

Images