Drug efficacy evaluation data analysis method based on multidimensional biochemical indicators
By constructing a discrete-time state estimation model and utilizing a robust weighting strategy, the problem of inaccurate drug efficacy assessment data caused by behavioral interference in live animal experiments was solved, achieving accurate drug efficacy assessment and improved data quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- KCI BIOTECH(SUZHOU) INC
- Filing Date
- 2026-03-12
- Publication Date
- 2026-07-17
AI Technical Summary
In cardiovascular drug development, existing technologies cannot effectively remove non-Gaussian pulse noise caused by behavioral interference in live animal experiments, leading to inaccurate drug efficacy evaluation data.
A drug efficacy evaluation data analysis method based on multidimensional biochemical indicators was adopted. By constructing a discrete-time state estimation model, an anti-differential weighting strategy was used to identify and suppress non-physiological mutation features, and efficacy evaluation was carried out in combination with the efficacy evaluation baseline.
This method enables the restoration and extraction of real physiological trends in a high-noise environment, improving the accuracy and reliability of drug efficacy evaluation and reducing experimental costs.
Smart Images
Figure CN121812207B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical bioinformatics data processing technology, specifically to a method for analyzing drug efficacy evaluation data based on multidimensional biochemical indicators. Background Technology
[0002] In cardiovascular drug development and pathological mechanism research, establishing stable and reliable animal models is a key step in evaluating drug efficacy, such as the DOCA salt-induced hypertensive heart failure rat model. Such experiments usually require long-term continuous monitoring of physiological parameters of the subjects, among which the trend of blood pressure change is the core indicator for judging whether the drug is effective.
[0003] However, in the actual process of medical information collection and processing, there is a technical pain point: the mixing of physiological signals and behavioral noise. Since rats are living animals, they are easily affected by environmental stimuli or their own emotions during non-invasive tail cuff pressure measurement or invasive intubation pressure measurement, which can cause them to struggle violently. This struggling behavior can cause irrational and violent fluctuations in blood pressure readings in a very short time, such as a sudden spike from 120 mmHg to 200 mmHg. From the perspective of signal processing, this fluctuation is a typical non-Gaussian impulse noise.
[0004] Existing technologies typically employ the following two processing methods:
[0005] Removal based on static thresholds involves setting a fixed upper limit, such as 180 mmHg, and values exceeding the upper limit will be deleted. However, as heart failure worsens, the rats' baseline blood pressure will rise, and static thresholds may lead to the accidental deletion of real pathological hypertension data, resulting in false negative assessments.
[0006] Conventional linear filtering methods, such as Kalman filtering or moving average filtering, assume that the noise follows a Gaussian distribution and adheres to the minimum mean square error criterion. When encountering large-amplitude impulse noise, in order to smooth out the error, the obtained estimated value will be forcibly dragged towards the direction of outliers, resulting in false peaks in the filtered curve and seriously interfering with the accurate calculation of the drug's blood pressure reduction effect. Summary of the Invention
[0007] To address the problem that existing technologies cannot accurately extract the true trends of physiological and biochemical indicators from complex signals containing high-intensity behavioral interference, this invention provides a drug efficacy evaluation data analysis method based on multidimensional biochemical indicators. This method includes:
[0008] Real-time multidimensional biochemical index sequences of the target organism during drug intervention are collected and preprocessed. A discrete-time state estimation model is constructed based on the real-time multidimensional biochemical index sequences to calculate the prior state estimate at the current sampling time. Simultaneously, the observation residual between the observed value at the current sampling time and the prior state estimate in the real-time multidimensional biochemical index sequences is calculated. Based on the statistical distribution characteristics of the observation residuals, a preset robust weighting strategy is used to calculate the signal confidence weight at the current sampling time. This robust weighting strategy is used to identify and suppress non-physiological mutation features in the real-time multidimensional biochemical index sequences. The observation error parameters of the discrete-time state estimation model are dynamically adjusted in real-time using the signal confidence weights to determine the gain matrix. A posterior state estimate is generated using the gain matrix, the observed value at the current sampling time, and the prior state estimate. The posterior state estimate is used as cleaned physiological trend data and combined with a preset drug efficacy evaluation baseline to perform drug efficacy evaluation, resulting in a drug efficacy score.
[0009] This invention effectively solves the technical problem of inaccurate drug efficacy evaluation data caused by stress-induced struggle of the test organism in live animal experiments, and realizes the restoration and extraction of the true physiological trend of the test organism in a strong noise environment.
[0010] Furthermore, the real-time multidimensional biochemical index sequence includes: systolic blood pressure, diastolic blood pressure, and instantaneous heart rate of the organism under test; the systolic blood pressure, diastolic blood pressure, and instantaneous heart rate at the same sampling time are constructed into a three-dimensional observation vector, which serves as the input unit of the real-time multidimensional biochemical index sequence.
[0011] This invention constructs a three-dimensional observation vector from the systolic blood pressure, diastolic blood pressure, and instantaneous heart rate of the organism under test as the input unit. Compared with traditional single-index analysis, it can capture the comprehensive effect of drugs at the hemodynamic level more comprehensively and three-dimensionally, avoiding one-sided misjudgment caused by occasional fluctuations of a single index.
[0012] Furthermore, the preprocessing includes: acquiring raw signals through a preset biomedical signal acquisition interface, performing analog-to-digital conversion on the raw signals to obtain a digital signal sequence; performing real-time feature analysis on the digital signal sequence to extract physiological indicators of the organism under test; and performing time-series alignment and multi-dimensional data fusion on the physiological indicators to construct a three-dimensional observation vector of the organism under test.
[0013] Furthermore, the observed residuals satisfy the following relationship:
[0014]
[0015] in, The observation residual; The observed value at the current sampling time; The observation matrix; The prior state estimate is given.
[0016] Furthermore, the robust weighting strategy includes: standardizing the observed residuals to obtain standardized residuals; setting a first threshold and a second threshold, wherein the first threshold is less than the second threshold; when the absolute value of the standardized residual is less than or equal to the first threshold, it is determined to be a normal physiological fluctuation, and the signal confidence weight is set to 1; when the absolute value of the standardized residual is greater than the first threshold and less than the second threshold, it is determined to be suspected interference, and the signal confidence weight decreases with a negative correlation as the standardized residual increases; when the absolute value of the standardized residual is greater than or equal to the second threshold, it is determined to be a non-physiological mutation, and the signal confidence weight is set to a preset minimum value.
[0017] This invention employs a three-stage weighting strategy based on robust estimation theory. By setting a grading threshold, the residual interval is divided into normal physiological fluctuations, suspected interference, and non-physiological mutations. This ensures that the algorithm maintains high sensitivity to normal small rhythms of the target organism, such as respiratory arrhythmia, while adaptively processing mechanical interference that exceeds statistical boundaries from softening and reducing weights to forced removal. This achieves the best balance between the robustness of the algorithm and signal fidelity.
[0018] Furthermore, the first threshold and the second threshold are set based on biostatistical principles.
[0019] Furthermore, the observation error parameter is specifically the observation noise covariance matrix of the discrete-time state estimation model. Real-time dynamic adjustment of the observation error parameter of the discrete-time state estimation model includes: when the signal confidence weight decreases due to the identification of non-physiological mutations in the organism under test, increasing the element values of the observation noise covariance matrix to reduce the gain matrix.
[0020] This invention establishes a negative correlation mapping mechanism between the observation noise covariance matrix and the signal confidence weight. When a non-physiological mutation is detected, the value of the observation error parameter is actively increased, and the gain matrix is reduced according to the minimum mean square error criterion. This physically cuts off the correction path of instantaneous high-energy noise on the posterior state estimate, thus preventing filter divergence.
[0021] Furthermore, the drug efficacy score satisfies the following relationship:
[0022]
[0023] in, The efficacy score of the drug is given; This is the sequence index at the sampling time; To assess the total number of sampling points corresponding to the deadline; This is the estimated posterior state value; The sampling time interval; This serves as the baseline for the efficacy evaluation. It is a drug metabolism attenuation factor; It is the Euclidean norm.
[0024] This invention constructs a pharmacodynamic scoring model that includes Euclidean distance calculation and drug metabolism decay factor. The pharmacodynamic deviation is characterized by calculating the Euclidean distance between the multidimensional state vector and the baseline vector. The model is then weighted and integrated by combining the time decay characteristics that conform to the pharmacokinetic law. The final drug efficacy score can objectively and accurately reflect the real metabolic process and cumulative effect of the drug in the body.
[0025] Furthermore, the drug metabolism attenuation factor was comprehensively calibrated through an offline pharmacokinetic experiment.
[0026] Furthermore, the efficacy evaluation baseline includes: collecting resting-state multidimensional biochemical index data of the test organism before drug administration, calculating the arithmetic mean of the multidimensional biochemical index data in each dimension, constructing a multidimensional mean vector, and using the multidimensional mean vector as the efficacy evaluation baseline.
[0027] The present invention has the following technical effects:
[0028] This invention constructs an anti-interference data analysis system that integrates biostatistics principles, modern control theory, and pharmacokinetics. Through a progressive processing logic of multi-dimensional feature fusion, adaptive robust filtering, and objective efficacy evaluation, it not only solves the common industry problem of difficult removal of behavioral noise from conscious experimental animals and restores the true pathological evolution and drug action curves, but also establishes a standardized and reproducible objective evaluation paradigm for drug screening. This effectively improves the data quality and experimental success rate of preclinical drug research, reduces experimental costs, and provides highly reliable data support for new drug development. Attached Figure Description
[0029] Figure 1 This is a flowchart of a drug efficacy evaluation data analysis method based on multidimensional biochemical indicators provided in an embodiment of the present invention;
[0030] Figure 2 This is a comparison diagram of the effects of the prior art and the method of the present invention provided in the embodiments of the present invention. Detailed Implementation
[0031] Before detailing the specific embodiments of the present invention, it should be noted that the biochemical indicators mentioned in the title and text of the present invention refer to a comprehensive indicator system that reflects the chemical changes and physical state of an organism's life activities. In the scenarios of cardiovascular disease drug development and heart failure model efficacy evaluation, since hemodynamic parameters such as systolic blood pressure and diastolic blood pressure are directly regulated by the hormone levels of the renin-angiotensin-aldosterone system in the body, blood pressure data is described as a core biochemical indicator in this embodiment, which is in line with the definition of processing medical and health data.
[0032] This invention provides a method for analyzing drug efficacy evaluation data based on multidimensional biochemical indicators, referring to... Figure 1 This includes steps S1-S4:
[0033] S1: Data Acquisition and Preprocessing.
[0034] Specifically, this step aims to establish a physical link between the organism under test and the digital analysis system. Through a high-precision signal conversion and cleaning mechanism, continuous physiological signals are transformed into time-synchronized, dimension-aligned discrete digital sequences, providing standardized input vectors for subsequent calculations.
[0035] In this embodiment, healthy SD rats were first selected as the basic experimental sample. In order to construct a test organism with stable pathological characteristics and drug sensitivity, this embodiment used deoxycorticosterone acetate combined with high-salt drinking water induction method to pathologically model the basic experimental sample and construct a live animal model with pathological characteristics of hypertension and heart failure, which was used as the test organism.
[0036] This embodiment uses a combination of a non-invasive tail-type blood pressure monitor and an implantable bioelectrode to acquire raw signals from the organism under test. Specifically, the organism under test is placed in a thermostatic fixation device, and a high-sensitivity pressure sensor is attached to the tail artery to acquire pulse signals in real time, obtaining a pulse signal sequence. The body surface electrocardiogram signals are acquired in real time through a subcutaneously implanted three-lead electrode to obtain an electrocardiogram signal sequence. All acquisition devices are connected to a preset biomedical signal acquisition interface.
[0037] In this embodiment, a PowerLab multi-channel physiological recorder is used as the core hardware for the biomedical signal acquisition interface. During the acquisition process, a low-pass digital filter with a cutoff frequency of 50Hz is used to filter the acquired raw signal in real time to remove power frequency interference and electromyography artifacts. Then, the filtered raw signal is converted from analog to digital at a high-frequency sampling frequency of 1kHz to ensure that subtle features such as the descending notch in the blood pressure waveform can be captured to obtain a digital signal sequence.
[0038] Real-time feature analysis of digital signal sequences yields specific physiological index values, as follows:
[0039] Peak and trough detection is performed on the pulse signal sequence to identify the highest point of the waveform within each cardiac cycle, and its amplitude is defined as the systolic blood pressure. For the first... The systolic pressure at each sampling time is denoted as... Identify the lowest point of the waveform within each cardiac cycle and define its amplitude as diastolic pressure. For the first... The diastolic blood pressure at each sampling time is denoted as... ;
[0040] R-wave peak detection is performed on the electrocardiogram signal sequence, the time interval between two adjacent R waves is calculated, and this time interval is converted into instantaneous heart rate. For the first R wave... The instantaneous heart rate at each sampling moment is denoted as . .
[0041] Since the collected data originates from different acquisition channels and processing algorithms, slight timing deviations may occur. Therefore, this embodiment adopts a timing alignment method based on hardware synchronization triggering. Using the system master clock as a reference, vector encapsulation technology is used to construct a three-dimensional observation vector from the three key indicators extracted at the same sampling time, denoted as . , Finally, a real-time multidimensional biochemical indicator sequence was obtained. .
[0042] It should be noted that the description of the organism to be tested, hardware device, and processing algorithm in this embodiment is only a preferred embodiment of the present invention. In practical applications, the basic experimental sample is not limited to SD rats, but can also be mice, rabbits, beagles, non-human primates, and other medical experimental animals; the pathological model is not limited to the deoxycorticosterone acetate combined with high-salt drinking water induction method, but can also be spontaneous hypertension, surgical ligation induction, or cardiovascular disease models constructed by gene editing; the biomedical signal acquisition interface is not limited to the PowerLab multi-channel physiological recorder, but any device with multi-channel biosignal synchronous acquisition function, such as the DIS implantable wireless telemetry system or the Biopac physiological signal acquisition system, can be used; real-time feature analysis can also use frequency domain analysis and wavelet transform to extract systolic blood pressure, diastolic blood pressure, and instantaneous heart rate, all of which should be included within the protection scope of the present invention.
[0043] S2: Construct a discrete-time state estimation model.
[0044] Specifically, this step is based on the discrete-time state estimation method in control theory to construct a mathematical benchmark model that can characterize the physiological changes of the organism under test in an ideal, undisturbed state. The physiological state of the organism under test is then inertially extrapolated based on historical data using this model. The extrapolation results are compared with the observations collected by the sensors to calculate the observation residuals, which serve as the core basis for subsequent judgments on whether there are non-physiological mutations.
[0045] Considering that the regulatory effect of drugs on the physiological state of the test organism is usually a gradual and cumulative process rather than an instantaneous step change, this embodiment defines the following discrete-time state estimation model rule based on the basic principles of Kalman filtering and combined with the real-time multidimensional biochemical index sequence obtained by S1:
[0046] The evolution rules of the true physiological state are defined. Since a high sampling frequency of 1kHz is used in S1, the physiological indicators of the organism under test have strong inertia within a very short sampling interval without strong external stimuli. Therefore, the model stipulates that the true physiological state at the next sampling moment should be a continuation of the inertia of the previous sampling moment, superimposed with the organism's inherent small physiological fluctuations, such as normal fluctuations caused by respiratory rhythm, i.e., process noise. To assess the range of these inherent physiological fluctuations, the process noise covariance matrix is pre-defined. To assess the statistical magnitude of the aforementioned minute physiological fluctuations, It is configured as a diagonal matrix, and the values on its diagonal correspond to the physiological statistical variances of the systolic blood pressure, diastolic blood pressure, and instantaneous heart rate of the organism under test in the resting state.
[0047] The data generation rules are defined, and the collected data are defined as observation values. Since the sensor directly measures the target index without the need for dimension conversion, the model stipulates that the observation values output by the sensor should be equivalent to the actual physiological state of the organism under test, and external measurement noise, i.e., observation noise, is superimposed. This observation noise includes both weak electronic background noise and the severe artifacts caused by the organism's struggle, which is the focus of this invention. To evaluate the sensor's measurement error, a benchmark observation noise covariance matrix is pre-set. , It is configured as a diagonal matrix, and the values on its diagonal correspond to the hardware measurement accuracy variances of systolic blood pressure, diastolic blood pressure, and instantaneous heart rate, which are determined by the technical specifications provided by the sensor manufacturer.
[0048] At the initial sampling time, i.e. The filter needs to be initialized. In this embodiment, the data within the first second of the real-time multidimensional biochemical index sequence output by S1 is selected. Since the sampling frequency is 1kHz, the first 1000 three-dimensional observation vectors are selected. The arithmetic mean of systolic blood pressure, diastolic blood pressure, and instantaneous heart rate are calculated respectively. The three arithmetic means are used to construct the initial posterior state estimate. Simultaneously, define and initialize the error covariance matrix, setting the initial error covariance matrix as the identity matrix;
[0049] At each sampling time, the processor first uses the optimal state determined at the previous sampling time, i.e., the posterior state estimate, to further predict the prior state estimate at the current sampling time using the state transition matrix. The specific relationship is as follows:
[0050]
[0051] in, For the first The posterior state estimate at each sampling time; The state transition matrix is set as an identity matrix in this embodiment to represent state retention; For the first The prior state estimate at each sampling time.
[0052] Subsequently, the processor acquires the observation values at the current sampling time, i.e., the three-dimensional observation vector. The observed residuals are calculated by comparing them with the prior state estimates. The specific relationship is as follows:
[0053]
[0054] in, For the first The observation residuals at each sampling time; For the first Observations at each sampling time; For the first The observation matrix at each sampling time point is set as an identity matrix in this embodiment, representing direct observation.
[0055] If calculated An increase in the value indicates that a non-physiological mutation that violates the physiological inertia of the organism under test may have occurred at the current sampling time.
[0056] At each sampling time, the processor uses the computational uncertainty of the previous sampling time, i.e., the posterior error covariance matrix, to perform a time update operation and calculate the prior error covariance matrix for the current sampling time. The specific relationship is as follows:
[0057]
[0058] in, For the first The prior error covariance matrix at each sampling time; For the first The posterior error covariance matrix at each sampling time; Let be the process noise covariance matrix.
[0059] It should be noted that the discrete-time state estimation model and Kalman filter framework constructed in this embodiment are only a preferred method for realizing the technical concept of this invention. In practical applications, the discrete-time state estimation model is not limited to the standard linear Kalman filter. Those skilled in the art can use recursive state estimation algorithms such as extended Kalman filter, unscented Kalman filter, H-infinity filter, and particle filter to predict and update the physiological state of the organism under test, all of which fall within the protection scope of this invention. The state transition matrix and observation matrix are also not limited to identity matrices. When the sampling frequency is reduced or there is a nonlinear mapping relationship between the physical quantity measured by the sensor and the state quantity, a non-identity matrix or Jacobian matrix can be constructed based on the specific physical model. The process noise covariance matrix and the benchmark observation noise covariance matrix can also be determined by Allen's analysis of variance, least squares fitting, and statistical learning methods based on a large amount of historical data. For the model initialization method, in addition to using the arithmetic mean, the median, mode, and fixed values based on the resting baseline can be used for initialization, all of which are core technologies of this invention.
[0060] S3: Confidence assessment and state update based on robust weighting.
[0061] Specifically, based on the observation residuals obtained in S2, the reliability of the collected data is evaluated in real time. If the observation residuals increase, it means that the data has changed abruptly and is judged as non-physiological interference. By inversely expanding the observation noise covariance matrix, the filter ignores the observation value at the current sampling time, thereby achieving intelligent shielding against abnormal artifacts.
[0062] Since systolic blood pressure, diastolic blood pressure, and instantaneous heart rate have different physical dimensions and numerical ranges, directly using observation residuals for judgment lacks statistical basis. Therefore, this embodiment introduces a priori error covariance matrix and a benchmark observation noise covariance matrix to calculate the theoretical covariance matrix. The specific relationship is as follows:
[0063]
[0064] in, For the first The theoretical covariance matrix at each sampling time point is obtained. The three elements corresponding to systolic blood pressure, diastolic blood pressure, and instantaneous heart rate on the diagonal of this matrix are extracted, and their square roots are calculated to obtain the result for each index at the [number] sampling time point. The theoretical standard deviation at each sampling time point is used to obtain the standardized residuals by dividing the values of these three indicators in the observed residuals by their corresponding theoretical standard deviations. This is denoted as... , representing the The number of times the data collected at each sampling time deviates from the predicted value.
[0065] Based on standardized residuals, a three-segment robust weighting function is used to set the robust weighting strategy. The three indicators in the standardized residuals are independently evaluated. For each indicator, its corresponding signal confidence weight is calculated using its own standardized residual value, as detailed below:
[0066]
[0067] in, Weights for signal confidence; As a preset minimum value, this embodiment takes 10. -6 ; The first threshold, The second threshold is used as the basis for determining the signal confidence weight. It can be seen that when the absolute value of the standardized residual of a certain indicator is less than or equal to the first threshold, it is judged as a normal physiological fluctuation, and the signal confidence weight is set to 1. When the absolute value of the standardized residual of a certain indicator is greater than the first threshold but less than the second threshold, it is judged as suspected interference, and the signal confidence weight decreases negatively with the increase of the standardized residual. When the absolute value of the standardized residual of a certain indicator is greater than or equal to the second threshold, it is judged as a non-physiological mutation, and the signal confidence weight is set to 0. This judgment process is independent, and each indicator does not affect the others. Ultimately, three independent signal confidence weights are obtained.
[0068] It should be noted that, and These settings are all based on principles of biostatistics. Specifically, the settings are based on the standard normal distribution theory and the Laida criterion in statistics, as follows:
[0069] Since the observed residuals are standardized by introducing a theoretical covariance matrix, theoretically, under ideal conditions without abnormal interference, the standardized residuals follow a normal distribution. According to the probability density function of the normal distribution, the probability that the random variable falls within the interval [-1.5, 1.5] is 86.64%. This indicates that the vast majority of normal physiological fluctuations and sensor background noise should be distributed within this range. Based on this statistical fact, this embodiment takes... This ensures that while filtering out noise, the true physiological details of the drug's onset of action can be preserved to the greatest extent possible.
[0070] Based on the principle of low probability events in the standard normal distribution, the probability of a random variable falling within the range [-3, 3] is as high as 99.73%. This indicates that values exceeding this range constitute extreme anomalies. Based on this statistical fact, this embodiment takes... This ensures the robustness of the algorithm.
[0071] The baseline observation noise covariance matrix is dynamically corrected in real time using three independent signal confidence weights. This correction follows the inverse expansion principle, dividing the elements on the diagonal of the baseline observation noise covariance matrix corresponding to systolic blood pressure, diastolic blood pressure, and instantaneous heart rate by their respective signal confidence weights to obtain the equivalent noise observation covariance matrix. For the ... The equivalent noise observation covariance matrix at each sampling time is denoted as . .
[0072] The processor utilizes the equivalent noise observation covariance matrix, combined with the prior state estimate and the prior error covariance matrix, to perform Kalman filtering for state updates, as follows:
[0073] The Kalman gain matrix is calculated as the gain matrix used to generate the posterior state estimate, and the specific relationship is as follows:
[0074]
[0075] in, For the first The gain matrix at each sampling time; For the first The prior error covariance matrix at each sampling time; For the first The observation matrix at each sampling time, for Transpose of; For the first The equivalent noise observation covariance matrix at each sampling time;
[0076] It can be seen that for systolic blood pressure, diastolic blood pressure, and instantaneous heart rate, regardless of which indicator experiences a decrease in signal confidence weight due to non-physiological mutations, according to the principle of inverse expansion, The corresponding element values will increase, and the calculated gain matrix will decrease;
[0077] Next, using the gain matrix, the observation value at the current sampling time, and the prior state estimate, the posterior state estimate is generated, with the specific relationship as follows:
[0078]
[0079] in, For the first The posterior state estimate at each sampling time; For the first Prior state estimates at each sampling time; This is the gain matrix; For the first Observations at each sampling time; For the first The observation matrix at each sampling time.
[0080] It can be seen that when When decreasing, Item, that is The reduction indicates that the influence weight of the observation residuals was reduced when generating the posterior state estimate, making the final result more dependent on the prior state estimate and effectively smoothing out non-physiological mutations.
[0081] The error covariance matrix is updated using the following formula:
[0082]
[0083] in: For the first The posterior error covariance matrix at each sampling time; It is the identity matrix; For the first The prior error covariance matrix at each sampling time.
[0084] S4: Generation of drug efficacy score.
[0085] Specifically, the efficacy evaluation baseline is calculated based on the resting-state multidimensional biochemical index data of the organism before drug administration. Then, the weighted Euclidean distance-time accumulation method is used to calculate the deviation of the posterior state estimate at each sampling time after drug administration from the efficacy evaluation baseline. The deviation is then weighted by combining the drug metabolism decay factor, and all weighted deviation values within the evaluation window are accumulated to obtain the drug efficacy score.
[0086] It should be noted that, to ensure the accuracy of efficacy evaluation and the consistency of data dimensions, resting-state multidimensional biochemical index data specifically refer to the stable time window before drug administration. In this embodiment, 10 minutes before drug administration is selected, and the posterior state estimates within this period are extracted. The arithmetic mean of each index is calculated to obtain the baseline for efficacy evaluation, denoted as . ;
[0087] By consulting the drug's instructions, and setting the evaluation window to three times the elimination half-life stated in the instructions, this embodiment sets the total evaluation window duration to 60 minutes, i.e., the evaluation cutoff time is 60 minutes. Based on the 1kHz sampling frequency set in S1, the total number of sampling points within this evaluation window is 3,600,000. Next, the drug efficacy score is calculated, with the specific formula as follows:
[0088]
[0089] in, To score the efficacy of the drug; This is the sequence index at the sampling time; To assess the total number of sampling points corresponding to the deadline; This is the posterior state estimate; The sampling time interval is given by S1. ; This serves as a baseline for efficacy evaluation; This is a drug metabolism attenuation factor, representing the rate of drug elimination through metabolism in the target organism. It is the Euclidean norm.
[0090] It can be seen that in the initial stage of drug administration, Smaller The term approaches 1, and as time goes on, Increase The exponential decrease in the number of items is consistent with the objective fact that drug concentration gradually decreases and drug efficacy gradually weakens.
[0091] It should be noted that the drug metabolism attenuation factor was determined through a comprehensive offline pharmacokinetic experiment, the specific steps of which are as follows:
[0092] Ten SPF-grade male SD rats, weighing between 200g and 20g, were selected. All experimental procedures complied with the requirements of the Laboratory Animal Ethics Committee. The SD rats were housed in independent ventilated cages, with a constant temperature of 22℃±2℃ and a constant humidity of 55%±5%, maintaining a standard circadian rhythm of 12 hours of light and 12 hours of darkness. The rats underwent a 3-day acclimatization period before the experiment, and were fasted for 12 hours before the experiment, with free access to water to eliminate interference from food on drug absorption and metabolism.
[0093] Weigh the standard drug to be tested, dissolve it in physiological saline or a special solvent, and prepare a homogeneous and transparent injection solution. Administer a single dose to 10 SD rats via intraperitoneal injection. To ensure that the effective drug concentration can still be detected in the blood sample in the later stage of metabolism, the dose is twice the standard treatment dose. In this example, the dose is 5 mg / kg. The moment the administration is completed is recorded as t=0.
[0094] We set up a dense-sparse-dense logarithmic distribution sampling point to focus on capturing the characteristics of the distribution phase (rapid descent) and the elimination phase (slow descent). The specific time set is as follows:
[0095]
[0096] The rat alternating tail vein micro-blood collection method was used, with 0.3 ml collected at each time point. Immediately after blood collection, the blood was injected into an EP tube pre-filled with heparin sodium and gently inverted 3-5 times to prevent coagulation. The collected blood samples were then placed in a low-temperature centrifuge at 4°C and centrifuged at 3500 rpm for 10 minutes. The upper light yellow plasma layer was aspirated, aliquoted into cryovials, and stored in an ultra-low temperature freezer at -80°C for later testing.
[0097] Take 50 μL of plasma sample and add 150 μL of acetonitrile solution containing internal standard to precipitate protein; vortex for 3 minutes, then centrifuge at 12000 rpm for 10 minutes and collect the supernatant; analyze the supernatant using liquid chromatography-tandem mass spectrometry (LC-MS / MS). A general-purpose C18 reversed-phase column was used, with gradient elution of the mobile phase: phase A was an aqueous solution containing 0.1% formic acid, and phase B was acetonitrile. The flow rate was set to 0.3 mL / min, and the column temperature was maintained at 40℃. An electrospray ionization source was used, and scanning was performed in positive ion mode. Multiple reaction monitoring (MRM) was used for detection. The blood drug concentration in each sample was calculated based on the detected peak area.
[0098] The arithmetic mean of the blood drug concentration values of all SD rats at the same sampling time point was calculated, and the mean drug-time curve was plotted. The elimination phase at the end of the mean drug-time curve was taken, and the elimination phase was linearized using a one-compartment model equation to obtain the natural logarithm of the blood drug concentration. A coordinate system was established with time as the horizontal axis and the natural logarithm of the blood drug concentration as the vertical axis. Linear regression was performed on the data points, and the absolute value of the slope of the fitted line was the value of the drug metabolism attenuation factor. In this example, it was set to 0.173.
[0099] It should be noted that the above-mentioned experiment for calibrating the drug metabolism attenuation factor is only a preferred embodiment of the present invention. In practical applications, this parameter can also be obtained by using professional pharmacokinetic analysis software to automatically fit and solve the blood drug concentration data; other pharmacokinetic models such as two-compartment models and non-compartment models can also be used for fitting; for marketed or known drugs with well-defined pharmacological characteristics, the drug metabolism attenuation factor can be directly calculated based on the half-life parameters recorded in authoritative medical literature, pharmacopoeias, and drug instructions, without the need to repeat animal experiments.
[0100] To determine whether the test drug produced a genuine pharmacological effect, rather than a stress response caused by the injection procedure, the processor executes the following decision logic:
[0101] First, a negative control baseline was constructed. Under the same environment, equipment, and algorithm parameter settings as the S1 drug administration experiment, another group of SD rats was selected as the negative control group. In this example, 6 SD rats were selected.
[0102] The SD rats in this group were injected with an equal volume of physiological saline without any active drug components. The drug efficacy score of the SD rats in this group was calculated and recorded as the blank score. The arithmetic mean and standard deviation of all blank scores were calculated, and the sum of the arithmetic mean and three times the standard deviation was set as the drug efficacy judgment threshold.
[0103] Calculated Compare with the efficacy determination threshold, if If the efficacy threshold is exceeded, the drug is deemed effective. If the value is less than or equal to the efficacy determination threshold, it is deemed invalid.
[0104] If deemed valid, according to The numerical values are classified into categories, if If the efficacy threshold is greater than but less than twice the efficacy threshold, it is judged as weakly effective; if A drug efficacy threshold that is twice or greater than the threshold value is considered potent.
[0105] Figure 2 This is a comparison diagram of the effects of the prior art and the method of the present invention provided in the embodiments of the present invention. As shown in the figure, when SD rats generate non-physiological blood pressure pulses due to operational stress, the prior art cannot identify this abnormality and is mistakenly pulled upwards, forming a significant false bulge. Furthermore, there is a significant regression lag after the blood pressure pulse disappears, resulting in the results obtained during this period being significantly higher than the true physiological baseline, causing false negative judgments in drug efficacy evaluation. In contrast, when the present invention detects a blood pressure pulse, it automatically reduces the signal confidence weight of the data to close to 0, and the results obtained always closely follow the true pathological trend, achieving accurate tracking of the true physiological baseline.
Claims
1. A method for analyzing drug efficacy evaluation data based on multidimensional biochemical indicators, characterized in that, include: Real-time multidimensional biochemical index sequences of the organism under test during drug intervention were collected and preprocessed. A discrete-time state estimation model is constructed based on the real-time multidimensional biochemical index sequence. The prior state estimate at the current sampling time is calculated, and the observation residual between the observed value at the current sampling time and the prior state estimate in the real-time multidimensional biochemical index sequence is calculated. Pre-set the baseline observation noise covariance matrix; Based on the statistical distribution characteristics of the observation residuals, the signal confidence weight at the current sampling time is calculated using a pre-set robust weighting strategy. The robust weighting strategy is used to identify and suppress non-physiological mutation features in the real-time multidimensional biochemical index sequence. The robust weighting strategy includes: standardizing the observed residuals to obtain standardized residuals; setting a first threshold and a second threshold, where the first threshold is less than the second threshold; when the absolute value of the standardized residual is less than or equal to the first threshold, it is judged as normal physiological fluctuation, and the signal confidence weight is set to 1; when the absolute value of the standardized residual is greater than the first threshold but less than the second threshold, it is judged as suspected interference, and the signal confidence weight decreases negatively with the increase of the standardized residual; when the absolute value of the standardized residual is greater than or equal to the second threshold, it is judged as non-physiological mutation, and the signal confidence weight is set to a preset minimum value; thus, three independent signal confidence weights are obtained. The benchmark observation noise covariance matrix is dynamically corrected in real time using the three independent signal confidence weights. The correction rule follows the inverse expansion principle, dividing the corresponding elements on the diagonal of the benchmark observation noise covariance matrix by their corresponding signal confidence weights to obtain the equivalent noise observation covariance matrix. The observation error parameters of the discrete-time state estimation model are dynamically adjusted in real time using signal confidence weights to determine the gain matrix. The posterior state estimate is generated using the gain matrix, the observed value at the current sampling time, and the prior state estimate. The posterior state estimate is used as the cleaned physiological trend data and combined with the preset drug efficacy evaluation baseline to evaluate the drug efficacy and obtain the drug efficacy score. The observation error parameter is specifically the observation noise covariance matrix of the discrete-time state estimation model. Real-time dynamic adjustment of the observation error parameter of the discrete-time state estimation model includes: when the signal confidence weight decreases due to the identification of non-physiological mutations in the organism under test, increasing the element values of the observation noise covariance matrix to reduce the gain matrix.
2. The method for analyzing drug efficacy evaluation data based on multidimensional biochemical indicators according to claim 1, characterized in that, The real-time multidimensional biochemical index sequence includes: systolic blood pressure, diastolic blood pressure, and instantaneous heart rate of the organism under test; the systolic blood pressure, diastolic blood pressure, and instantaneous heart rate at the same sampling time are constructed into a three-dimensional observation vector, which serves as the input unit of the real-time multidimensional biochemical index sequence.
3. The method for analyzing drug efficacy evaluation data based on multidimensional biochemical indicators according to claim 1, characterized in that, The preprocessing includes: acquiring raw signals through a preset biomedical signal acquisition interface; performing analog-to-digital conversion on the raw signals to obtain a digital signal sequence; performing real-time feature analysis on the digital signal sequence to extract physiological indicators of the organism under test; and performing time-series alignment and multi-dimensional data fusion on the physiological indicators to construct a three-dimensional observation vector of the organism under test.
4. The method for analyzing drug efficacy evaluation data based on multidimensional biochemical indicators according to claim 1, characterized in that, The observed residuals satisfy the following relationship: in, The observation residual; The observed value at the current sampling time; The observation matrix; The prior state estimate is given.
5. The method for analyzing drug efficacy evaluation data based on multidimensional biochemical indicators according to claim 1, characterized in that, The first threshold and the second threshold are set based on biostatistical principles.
6. The method for analyzing drug efficacy evaluation data based on multidimensional biochemical indicators according to claim 1, characterized in that, The drug efficacy score satisfies the following relationship: in, The efficacy score of the drug is given; This is the sequence index at the sampling time; To assess the total number of sampling points corresponding to the deadline; This is the estimated posterior state value; The sampling time interval; This serves as the baseline for the efficacy evaluation. It is a drug metabolism attenuation factor; It is the Euclidean norm.
7. The method for analyzing drug efficacy evaluation data based on multidimensional biochemical indicators according to claim 6, characterized in that, The drug metabolism attenuation factor was determined through a comprehensive offline pharmacokinetic experiment.
8. The method for analyzing drug efficacy evaluation data based on multidimensional biochemical indicators according to claim 1, characterized in that, The efficacy evaluation baseline includes: collecting resting-state multidimensional biochemical index data of the test organism before drug administration, calculating the arithmetic mean of the multidimensional biochemical index data in each dimension, constructing a multidimensional mean vector, and using the multidimensional mean vector as the efficacy evaluation baseline.