Clinical trial data prediction method based on big data

Abstract

The present invention relates to the field of electronic digital data processing technology, and specifically to a clinical trial data prediction method based on big data, comprising: obtaining the test data to be tested, historical test data at different times, and labels of each patient in the clinical trial, wherein the historical test data and the test data to be tested include feature data of multiple dimensions; taking the historical clinical trial data of any patient at different times as a sample, obtaining the average frequency of the IMF components of each sample on each dimensional feature data, and dividing them into low-frequency, medium-frequency, and high-frequency scale components according to the frequency range; calculating the probability that the test data to be tested belongs to each label, combining the discrimination of each scale component and the label probability output by the network model to obtain a main label, and the main label is the prediction result of the test data to be tested. The present invention solves the problem of low accuracy in predicting clinical trial results for patients.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and more particularly to a clinical trial data prediction method based on big data. Background Art

[0002] With the rapid advancement of medicine, clinical trials play a vital role in the development of new drugs, new therapies, and medical devices. Clinical trials not only provide a basis for the marketing of new drugs but also play an important role in disease prevention and treatment, efficacy evaluation, and drug safety monitoring. However, traditional clinical trials face many challenges in data collection, analysis, and prediction, especially in the processing and analysis of trial data. Traditional clinical trials often rely on relatively simple data analysis methods, such as linear regression and chi-square tests. Although these methods can provide some reference value in some cases, they often fail to fully and accurately reflect the diversity of patients and the differences in treatment effects when faced with complex and multidimensional data.

[0003] With the rapid development of big data and artificial intelligence technologies, the volume and variety of clinical trial data has exploded. At the same time, the complexity and high dimensionality of clinical trial data have increased the difficulty of data prediction. In the context of big data, extracting valuable information from massive amounts of data and establishing scientifically sound prediction models have become major challenges in clinical trial research. For example, when predicting drug efficacy, differences in patient genetic information, lifestyle, medical history, and other factors can lead to different responses to the same drug. Therefore, accurately predicting a patient's response to a drug and its efficacy is a critical issue in clinical trials.

[0004] However, despite the enormous potential of big data and machine learning technologies in clinical trials, their practical application still faces a series of challenges, such as data quality, data privacy protection, and model generalization. Traditional clinical trials often adopt a one-size-fits-all approach, ignoring individual differences among patients. This can lead to different responses to the same drug, resulting in inaccurate predictions of clinical trial results. Summary of the Invention

[0005] In order to solve the problem of low accuracy in predicting clinical trial results for patients raised in the above background technology, the present invention provides the following solution.

[0006] The present invention provides a clinical trial data prediction method based on big data, comprising: obtaining test data to be tested, historical test data at different times, and labels for each patient in the clinical trial, wherein the historical test data and the test data to be tested include feature data of multiple dimensions; taking the historical clinical trial data of any patient at different times as a sample, obtaining the average frequency of the IMF component of each sample on the feature data of each dimension, and taking the average frequency less than a first threshold as the low-frequency scale component, the average frequency greater than a second threshold as the high-frequency scale component, and the average frequency greater than the first threshold and less than the second threshold as the medium-frequency scale component; obtaining the probability that the test data to be tested belongs to each label, wherein the probability is the product of the probability of the label output by a network model based on each scale component and the discrimination of the label by the corresponding scale component; the discrimination is the mean of the sum of the balance degrees of any two types of label combinations in the historical test data on the corresponding scale component; the balance degree is positively correlated with the difference degree of the samples corresponding to the two types of label combinations on the corresponding scale component; the difference degree represents the degree of similarity between any two types of label combinations on the corresponding scale component; and taking the label with the highest probability as the main label of the test data to be tested, and the main label is the prediction result of the test data to be tested.

[0007] The above technical solution processes the average frequency of the IMF components of the sample in each dimension of feature data, and then effectively distinguishes the feature information of components of different scales, thereby improving the prediction accuracy of various labels and solving the problem of low accuracy in predicting patients' clinical trial results.

[0008] Furthermore, the network model is a long short-term memory network model.

[0009] The above technical solution, by using a long-short-term memory (LSTM) network model as a network model, can effectively process clinical trial data with time series characteristics. The LSTM network model has superior memory capacity and can capture long-term dependencies in data, making it particularly suitable for processing historical patient trial data and its changing trends. The LSTM network model's time series modeling capabilities enable precise analysis of changes in patient data between different time points, thereby improving the accuracy of label predictions. Combining multi-dimensional feature data with frequency analysis, the LSTM network model can more effectively explore potential nonlinear relationships, enhance the model's adaptability and predictive capabilities in complex data, and thus provide more reliable decision support for personalized treatment.

[0010] Furthermore, it also includes training the network model, specifically: inputting the training set into the pre-built network model for training, and during the training process, calculating the loss between the output prediction value and the label; adjusting the network model parameters using the gradient descent method to minimize the prediction error; iteratively adjusting the parameters of the network model until the loss is less than a certain value or reaches the set number of training times, and finally obtaining a trained network model.

[0011] Furthermore, the characteristic data of multiple dimensions include: heart rate, blood pressure, respiratory rate, body temperature, blood sugar, blood lipids and white blood cell count.

[0012] The above technical solution helps to comprehensively reflect the patient's physiological status and disease characteristics by incorporating characteristic data from multiple dimensions such as heart rate, blood pressure, respiratory rate, body temperature, blood sugar, blood lipids and white blood cell count into the analysis, enhances the model's ability to express clinical trial data, improves the accuracy and reliability of label predictions, makes treatment decisions more precise and personalized, and effectively supports clinical auxiliary diagnosis and treatment plan optimization.

[0013] Furthermore, the labels include: worsening, stable and improved.

[0014] Furthermore, it also includes denoising and standardizing the historical test data and the test data to be detected.

[0015] By denoising and standardizing historical and pending test data, this technical solution effectively removes noise and inconsistencies, ensuring data quality and stability. This process improves data comparability and enables more accurate subsequent analysis and predictions. Denoising eliminates the impact of outliers, while standardization ensures consistency across different data dimensions, thereby enhancing model training effectiveness and predictive capabilities, ultimately improving the reliability and accuracy of clinical trial data analysis.

[0016] Furthermore, the combination of any two types of labels includes three combinations: deterioration and stability, deterioration and improvement, and stability and improvement.

[0017] Furthermore, the average frequency of the IMF component is obtained by using an empirical mode decomposition or an ensemble empirical mode decomposition algorithm.

[0018] By employing empirical mode decomposition (EMD) or ensemble EMD algorithms to obtain the average frequency of IMF components, this technical solution can accurately decompose complex signals into intrinsic mode functions (IMFs) of varying scales, thereby revealing the multi-layered features hidden within the data. This effectively extracts the essential frequency characteristics of the data, making the role of different frequency components in data analysis more clearly understood, helping to improve the model's ability to analyze and predict complex clinical trial data. This decomposition method can better handle nonlinear and non-stationary data, enhancing the flexibility and accuracy of data processing and providing a more reliable foundation for subsequent label prediction.

[0019] Furthermore, the difference is the sum of the ratios of the absolute values ​​of the differences of the elements of the feature data of any two samples on the IMF components of the corresponding scale components to the maximum values ​​on the IMF components of the corresponding scale components.

[0020] Furthermore, the balance degree is the mean of the sum of the differences of the samples corresponding to the arbitrary two types of label combinations on the corresponding scale components.

[0021] The beneficial effects of the present invention are:

[0022] This paper combines big data analysis with advanced signal processing techniques to propose a precise prediction method based on multidimensional clinical trial data. By denoising and standardizing historical and pending test data, and extracting the signal's IMF components using an empirical mode decomposition algorithm, the model reveals distinct frequency characteristics in the data, enhancing its adaptability to complex clinical data. By quantifying the degree of differentiation between different labels and combining data learning and training with a network model, it is possible to accurately predict a patient's primary label and develop personalized treatment plans. This improves the accuracy and reliability of clinical trial data analysis, contributing to improved treatment outcomes, optimized patient management, and the implementation of personalized medicine. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 FIG. 4 is a flowchart schematically illustrating a clinical trial data prediction method based on big data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0024] An embodiment of a clinical trial data prediction method based on big data.

[0025] like Figure 1 As shown in FIG, a flow chart of a clinical trial data prediction method based on big data according to an embodiment of the present invention includes the following steps:

[0026] S1: Obtain the test data to be tested for each patient in the clinical trial, the historical test data at different times, and the corresponding labels.

[0027] In one embodiment, the historical test data and the test data to be detected include feature data of multiple dimensions, which are derived from clinical measurement records of patients at different time points and cover multiple core physiological indicators to ensure data integrity.

[0028] Feature data in multiple dimensions, including heart rate, blood pressure, respiratory rate, body temperature, blood sugar, blood lipids, and white blood cell count;

[0029] The labels include: worsening, stable, and improved. Worsening means that the patient's condition has worsened, stable means that the patient's condition has not changed significantly, and improved means that the patient's condition has recovered.

[0030] Furthermore, before conducting in-depth analysis of historical test data and untested test data, data preprocessing, including noise removal and normalization, is essential. This preprocessing step is crucial for ensuring data quality and improving the accuracy and stability of subsequent model analysis.

[0031] First, data denoising is a key step in eliminating outliers and sudden noise in the data. To achieve this goal, the present invention employs a median filtering method. Median filtering is a nonlinear filtering technique whose basic principle is to effectively suppress noise by replacing the value of each data point with the median of all data points in its neighborhood, particularly capable of removing spike noise or abnormal fluctuations. Compared to traditional mean filtering, median filtering is more effective in suppressing sudden noise, can preserve the boundary information and key features of the data, and is particularly suitable for sudden errors or abnormal data that may occur in clinical trial data.

[0032] Secondly, the purpose of standardization is to unify the scales of data of different dimensions so that they can be compared and analyzed at the same scale. In the present invention, the maximum and minimum value standardization method is adopted. This method scales all feature data to the same range by subtracting the minimum value of the feature from the data value of each feature, and then dividing it by the difference between the maximum and minimum values ​​of the feature. Compared with other standardization methods, the advantage of the maximum and minimum value method is that it does not change the distribution form of the data, but retains the relative relationship between each feature, avoids excessive interference with the data distribution, and makes the comparison of data between different dimensions more fair and consistent. Especially in clinical trial data, which involves feature data of multiple dimensions, standardization ensures that each parameter can be effectively compared and comprehensively evaluated under the same quantitative standard.

[0033] By combining these two steps—median filtering for denoising and minimum and maximum normalization—the quality of both historical and untested test data has been significantly improved. Outliers are effectively removed, and all feature data is brought to the same scale, laying a solid foundation for subsequent analysis, feature extraction, and predictive model training.

[0034] S2: Take the historical clinical trial data of any patient at different times as a sample, obtain the average frequency of the IMF component of each sample on each dimensional feature data, and divide it into low-frequency, medium-frequency and high-frequency scale components according to the first threshold and the second threshold.

[0035] In one embodiment, the average frequency of the IMF component is obtained by using an empirical mode decomposition or an ensemble empirical mode decomposition algorithm, and the average frequency less than the first threshold is used as a low-frequency scale component, greater than the second threshold is used as a high-frequency scale component, and greater than the first threshold and less than the second threshold is used as a medium-frequency scale component. The first threshold and the second threshold are 0.3 and 0.7 respectively, and of course they can also be set according to actual conditions; the patient's clinical trial data is adaptively decomposed by the empirical mode decomposition or the ensemble empirical mode decomposition algorithm, thereby effectively extracting different frequency components in the data, making the characteristics of the data at different scales clearer and easier to analyze. By setting two thresholds, the components are divided into low-frequency, medium-frequency and high-frequency scales, thereby achieving a hierarchical analysis of the long-term trend of the disease, short-term fluctuations and sudden abnormal conditions, thereby improving the interpretability and analysis accuracy of the data.

[0036] S3: Calculate the probability that the test data to be tested belongs to each label based on the probability of the network model outputting labels of different scale components and the discrimination degree of each scale component to the label.

[0037] In one embodiment, the probability is the product of the probability of the network model outputting the label based on each scale component and the discrimination of the label corresponding to each scale component; illustratively:

[0038] Calculate the probability that the test data to be tested belongs to each label , , where Based on scale component The probability of the network model output label, is the scale component Regarding the discrimination of labels; for, , The historical test data Class labels are combined in scale components The balance degree on the above, the label combinations include: deterioration and stability, deterioration and improvement, and stability and improvement;

[0039] By combining the characteristics of multiple scale components, we perform a weighted calculation of the probability of the label to which the test data belongs, thereby comprehensively evaluating the matching degree between patient data and different labels on a multi-dimensional basis. By introducing the discriminability of each scale component for labels, we can effectively capture the differential characteristics of the data at different frequency scales, thereby improving the accuracy of label prediction.

[0040] The balance is the mean of the sum of the differences of the samples corresponding to the arbitrary two-category label combination on the corresponding scale components; exemplarily:

[0041] The scale components of each label combination in the historical test data Balance for, , where Samples with the first label and samples of the second label In scale component The difference in is the number of samples of the first category label, is the number of samples of the second type of label, is the normalization function;

[0042] By calculating the average difference between samples with different labels at each scale component and normalizing the results, the discrimination at different scales can be compared within the same numerical range. Normalization not only eliminates data distribution imbalances but also enhances the comparability of components at different scales during the calculation process, preventing the excessive impact of certain scales on the overall calculation results due to their large numerical ranges. By calculating the average difference between all samples, we can more comprehensively measure the differences in feature distribution between different labels, reduce the interference of individual abnormal samples on the overall discrimination calculation, and improve the robustness and reliability of the data.

[0043] The difference is the sum of the ratios of the absolute values ​​of the differences between the elements of the feature data of any two samples on the IMF components of the corresponding scale components to the maximum values ​​on the IMF components of the corresponding scale components; exemplarily:

[0044] The difference is, , where For the and Samples in the weight The difference in For the Feature data of samples In scale component The first IMF component elements, For the Feature data of samples In scale component The first IMF component elements, To take the scale component The maximum value of the upper IMF component, For the Feature data in samples the number of Indicates the number of elements of the IMF component.

[0045] By comparing the feature data of samples with different labels at each scale component element by element and normalizing the difference values, the calculated difference can adapt to the feature distribution at different scales, enhancing data comparability. By traversing all feature data and IMF component elements, it is possible to comprehensively evaluate the subtle changes of different samples at each scale, ensuring the refinement of discrimination calculation and improving classification accuracy.

[0046] The network model is a long short-term memory network model.

[0047] Furthermore, the network model is trained, specifically by: obtaining a training set, wherein the training set includes the IMF components of the characteristic data of each sample in the historical test data at different scale components; wherein the labels are set manually;

[0048] The training set is input into the pre-built network model for training. During the training process, the loss between the output prediction value and the label is calculated; the network model parameters are adjusted using the gradient descent method to minimize the prediction error; the network model parameters are iteratively adjusted until the loss is less than a certain value or the set number of training times is reached, and finally a trained network model is obtained.

[0049] S4: The label with the highest probability is used as the main label of the test data to be detected. The main label is the prediction result of the test data to be detected.

[0050] In one embodiment, the label with the highest probability is selected as the primary label for the test data to be tested. This primary label is the predicted result of the test data to ensure the highest credibility of the label judgment, reduce the possibility of misjudgment, and make the label judgment more stable. By associating and matching the optimal treatment plan corresponding to different category labels, it is possible to reduce uncertainty in clinical decision-making, improve the targetedness and effectiveness of treatment, and thus provide patients with more scientific and reasonable treatment plans and improve overall treatment effectiveness.

[0051] The solution of the present invention provides an accurate clinical trial data prediction method by combining big data technology, frequency component analysis, network model training, and multi-dimensional feature data processing. This method can extract valuable features from complex historical test data and test data to be tested. It uses multi-level frequency analysis and difference calculation to enhance the ability to distinguish different labels. At the same time, denoising and normalization ensure data stability and consistency, and the training and optimization of the network model further improves prediction accuracy. It provides a reliable prediction basis for clinical decision-making, promotes the development of personalized medicine, and improves treatment efficacy and patient safety.

[0052] In the description of this specification, "multiple" and "several" mean at least two, such as two, three or more, etc., unless otherwise clearly defined.

[0053] While several embodiments of the present invention have been shown and described herein, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that various alternatives to the embodiments of the present invention described herein may be employed in practicing the present invention.

Claims

1. A clinical trial data prediction method based on big data, characterized in that: include: Obtaining the test data to be tested, historical test data at different times, and labels for each patient in the clinical trial. The labels include: deterioration, stability, and improvement. The historical test data and test data to be tested include feature data in multiple dimensions. Take the historical clinical trial data of any patient at different times as a sample, obtain the average frequency of the IMF component of each sample on each dimension feature data, and take the average frequency less than the first threshold as the low-frequency scale component, the average frequency greater than the second threshold as the high-frequency scale component, and the average frequency greater than the first threshold and less than the second threshold as the medium-frequency scale component; Obtain the probability that the test data to be tested belongs to each label. The probability is the product of the probability of the network model outputting the label based on each scale component and the discrimination degree of the label for the corresponding scale component; the discrimination degree is the mean of the sum of the balance degrees of any two types of label combinations in the historical test data on the corresponding scale component. Distinction of labels for, , The historical test data Class labels are combined in scale components The balance degree on the above, the label combinations include: deterioration and stability, deterioration and improvement, and stability and improvement; The balance degree is positively correlated with the difference between the samples corresponding to the two types of label combinations on the corresponding scale components; the balance degree is positively correlated with the difference between the samples corresponding to the two types of label combinations on the corresponding scale components in the historical test data. Balance for, , where Samples with the first label and samples of the second label In scale component The difference in is the number of samples of the first category label, is the number of samples of the second type of label, is the normalization function; The difference degree represents the similarity between any two types of label combinations on the corresponding scale components; the difference degree is, , where For the and Samples in the weight The difference in For the Feature data of samples In scale component The first IMF component elements, For the Feature data of samples In scale component The first IMF component elements, To take the scale component The maximum value of the upper IMF component, For the Feature data in samples the number of Indicates the number of elements of the IMF component; The label with the highest probability is used as the main label of the test data to be detected, and the main label is the prediction result of the test data to be detected.

2. The clinical trial data prediction method based on big data according to claim 1, characterized in that: The network model is a long short-term memory network model.

3. The clinical trial data prediction method based on big data according to claim 1, characterized in that: It also includes training the network model, specifically: The training set is input into the pre-built network model for training. During the training process, the loss between the output prediction value and the label is calculated; Use gradient descent to adjust network model parameters to minimize prediction error; Iteratively adjust the parameters of the network model until the loss is less than a certain value or the set number of training times is reached, and finally a trained network model is obtained.

4. The clinical trial data prediction method based on big data according to claim 1, characterized in that: The characteristic data of the multiple dimensions include: heart rate, blood pressure, respiratory rate, body temperature, blood sugar, blood lipids and white blood cell count.

5. The clinical trial data prediction method based on big data according to claim 1, characterized in that: It also includes denoising and standardizing the historical test data and the test data to be detected.

6. The clinical trial data prediction method based on big data according to claim 1, characterized in that: The average frequency of the IMF component is obtained by using an empirical mode decomposition or an ensemble empirical mode decomposition algorithm.

Images