An anti-tumor tablet stability prediction method based on multi-source data fusion

Abstract

The application discloses a kind of anti-tumor tablet stability prediction methods based on multi-source data fusion, specifically related to the field of pharmaceutical preparation stability prediction, including S1, data acquisition;S2, multidimensional feature parameter extraction;S3, stability prediction benchmark model construction;S4, real-time attenuation index calculation;S5, deviation threshold range determination;S6, stability deviation index analysis;S7, prediction result dynamic update;The application integrates multi-source data, constructs stability prediction benchmark model, is not limited to the simple trend analysis of single factor, realizes the accurate prediction of content attenuation, related substance growth, disintegration time limit change index by key influence factor marker, critical threshold setting and theoretical attenuation curve fitting, effectively improves the accuracy and timeliness of anti-tumor tablet stability prediction, provides scientific basis for drug expiration date evaluation and quality risk control.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of pharmaceutical preparation stability prediction, more particularly, the present application relates to an anti-tumor tablet stability prediction method based on multi-source data fusion. BACKGROUND

[0002] Under the development trend of digital quality control of pharmaceutical preparations, the stability guarantee of anti-tumor tablet is crucial to drug safety, and its stability is influenced by multi-dimensional factors such as raw material characteristics, production process, storage environment, etc. The whole life cycle data presents the characteristics of multi-source and large scale. In the existing anti-tumor tablet stability prediction technology, the empirical formula method based on accelerated test depends on the extrapolation of idealized conditions, ignoring the batch difference of raw materials and process fluctuations. Single factor regression analysis only focuses on the correlation of local parameters, and it is difficult to capture the synergistic effect of multiple factors. Traditional periodic sampling detection has data lag, and cannot reflect the dynamic changes of stability in real time.

[0003] These technical defects lead to significant limitations in actual application: first, multi-source data is not systematically integrated, raw material characteristics data, production process data, storage environment data and historical stability detection data are scattered in different systems, the data format is not unified, the correlation is not mined, and the complete data support cannot be provided for model construction, resulting in single and one-sided prediction dimension; second, there is no dynamic deviation correction mechanism, the benchmark model is fixed after construction, and the prediction logic cannot be adjusted combined with the current storage environment data and quality detection data monitored in real time. When the actual decay index deviates greatly from the theoretical benchmark, the initial model is still used for prediction, which may cause distortion of stability evaluation;

[0004] Third, the key parameter setting lacks flexibility, once the key influence factor weight, stability critical threshold and theoretical decay curve parameter are determined, they are fixed for a long time, and cannot respond to the stability law changes caused by raw material fluctuations, process optimization or storage condition changes, resulting in disconnection between prediction results and actual quality state, and difficulty in meeting the precise needs of anti-tumor tablet whole life cycle quality control. SUMMARY

[0005] Therefore, the embodiments of the present application provide an anti-tumor tablet stability prediction method based on multi-source data fusion. In order to achieve the above purpose, the present application provides the following technical scheme:

[0006] S1, a multi-source data acquisition module acquires raw data of multiple batches of anti-tumor tablets, and uploads the raw data to a tablet stability database;

[0007] S2, a data fusion processing module extracts multi-dimensional characteristic parameters of the same batch of tablets from the tablet stability database according to batch, and transmits the multi-dimensional characteristic parameters to a stability model construction module;

[0008] S3, the stability model construction module constructs a tablet stability prediction benchmark model according to the extracted multi-dimensional characteristic parameters, marks key influence factors, a stability critical threshold and a theoretical attenuation curve;

[0009] S4, the real-time monitoring data acquisition module acquires current storage environment data and real-time quality detection data of the tablet batch to be predicted, and transmits the data to the data fusion processing module to calculate real-time attenuation indexes of various stability indexes under the current condition;

[0010] S5, the stability deviation calculation module compares the real-time attenuation indexes with the theoretical attenuation benchmark in the stability prediction benchmark model, and calculates the deviation threshold range of each index;

[0011] S6, the stability deviation index analysis module comprehensively analyzes the stability deviation index based on the deviation threshold range of multiple indexes, and measures the deviation degree of the current tablet state from the theoretical model;

[0012] S7, the prediction result generation and dynamic updating module generates a stability prediction curve according to the stability deviation index, triggers a model dynamic optimization algorithm if the deviation exceeds the threshold, and generates a stability prediction curve again, or performs local correction based on the benchmark model if the deviation is small.

[0013] The technical effects and advantages of the present application are as follows:

[0014] The present application integrates whole-chain data such as raw material characteristics, production process, storage environment and historical stability detection through a multi-source data acquisition module, extracts multi-dimensional characteristic parameters through a data fusion processing module, is not limited to a single data dimension or simple statistical analysis, can provide complete data support for stability model construction on the one hand, and comprehensively covers the core factors affecting tablet stability, on the other hand, avoids prediction deviation caused by one-sided data through multi-source data correlation analysis, and improves the accuracy of stability law capture;

[0015] The stability model construction module of the present application marks key influence factors, a stability critical threshold and a theoretical attenuation curve, combines current environment and quality data obtained by the real-time monitoring data acquisition module, calculates real-time attenuation indexes through the data fusion processing module, quantifies the deviation degree through the stability deviation calculation module and the deviation index analysis module, triggers a model dynamic optimization algorithm to generate a prediction curve again if the deviation exceeds the threshold, or performs local correction if the deviation is small; this dynamic updating mechanism can not only optimize parameters to improve the fitting degree of the prediction curve and the actual state, but also quickly responds to environmental fluctuations through local correction, ensures that the prediction result is real-time adapted to the quality change of the tablet, avoids long-term prediction distortion caused by model staticization, provides precise data support for whole-life-cycle quality control;

[0016] The application is based on the comprehensive calculation of stability deviation index based on the deviation threshold range of multiple indexes, the influence of key quality attributes is strengthened through the differentiated setting of index weight, the quantitative evaluation of the degree of stability deviation is realized, potential quality risks can be accurately identified, through the hierarchical processing mechanism of dynamic optimization and local correction, resource waste caused by excessive adjustment is avoided, model maintenance cost is reduced, and the scientificity and efficiency of antitumor tablet quality control are improved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The overall structure flowchart of the application.

[0018] Figure 2 The overall module schematic diagram of the application. DETAILED DESCRIPTION

[0019] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0020] As shown in the accompanying drawings Figure 1 An antitumor tablet stability prediction method based on multi-source data fusion, comprising: a multi-source data acquisition module, a tablet stability database, a data fusion processing module, a stability model construction module, a real-time monitoring data acquisition module, a stability deviation calculation module, a stability deviation index analysis module, and a prediction result generation and dynamic updating module.

[0021] The multi-source data acquisition module is used to acquire original data of multiple batches of antitumor tablets.

[0022] The tablet stability database is used to store the original data of multiple batches of antitumor tablets acquired by the multi-source data acquisition module.

[0023] The data fusion processing module is used to extract multi-dimensional feature parameters from the tablet stability database and calculate real-time attenuation indexes of various stability indexes.

[0024] The stability model construction module is used to construct a tablet stability prediction benchmark model.

[0025] The real-time monitoring data acquisition module is used to acquire current data of tablets to be predicted in real time.

[0026] The stability deviation calculation module is used to calculate the deviation threshold range of each index.

[0027] The stability deviation index analysis module is used to obtain the stability deviation index;

[0028] The prediction result generation and dynamic update module is used to determine whether to regenerate the stability prediction curve.

[0029] As attached Figure 2 The method for predicting the stability of antitumor drug tablets based on multi-source data fusion, as shown, includes the following steps in its specific implementation:

[0030] S1. The multi-source data acquisition module collects raw data from multiple batches of antitumor drug tablets and uploads the raw data to the tablet stability database.

[0031] In this embodiment, it should be specifically stated that the original data includes: raw material characteristic data, production process data, storage environment data, and historical stability test data; raw material characteristic data includes the content of active ingredient in the active pharmaceutical ingredient, purity of excipients, particle size distribution, crystal structure, moisture content, and density; production process data includes granulation parameters: stirring speed and granulation time; tableting parameters: tableting pressure and tableting speed; drying parameters: temperature and time; coating parameters: thickness and curing time; storage environment data includes temperature, relative humidity, light intensity, oxygen concentration, and vibration frequency; historical stability test data includes content changes, related substance formation, disintegration time, dissolution rate, hardness, and friability under different storage durations.

[0032] It should be further explained that the raw material characteristic data are mainly collected through laboratory precision instrument testing methods. The content of active ingredients in the active pharmaceutical ingredient and the purity of excipients are determined by high performance liquid chromatography or gas chromatography-mass spectrometry. The particle size distribution of the raw materials is detected by a laser particle size analyzer. The crystal structure is analyzed by X-ray diffraction. The moisture content is determined by a Karl Fischer moisture analyzer. The density is obtained by a tap density meter or a specific gravity bottle method. The production process data are mainly collected through the built-in sensors of the production equipment and the industrial control system. Parameters such as granulation stirring speed and time, tableting pressure and speed, drying temperature and time are captured in real time by the equipment sensors and transmitted to the tablet stability database via industrial Ethernet. The coating thickness is detected by an online laser thickness gauge, and the curing time is recorded by a process timer. The storage environment data are mainly collected through a distributed environmental sensor network. Temperature and relative humidity are monitored by an integrated temperature and humidity sensor. Light intensity is collected by a photoelectric sensor. Oxygen concentration is detected by an electrochemical oxygen sensor. Vibration frequency is captured by an accelerometer. All environmental data are transmitted to the tablet stability database in real time via a wireless sensor network.

[0033] S2. The data fusion processing module extracts multi-dimensional feature parameters of tablets from the same batch from the tablet stability database and transmits them to the stability model construction module.

[0034] In this embodiment, it needs to be specifically pointed out that the multi-dimensional characteristic parameters specifically refer to the active ingredient content of the raw material, the purity of the auxiliary material, the particle size distribution of the raw material, the crystal structure, the moisture content, the density of the raw material characteristic dimension; The stirring speed of the granulation, the granulation time, the tabletting pressure, the tabletting speed, the drying temperature, the drying time, the coating thickness, the coating solidification time of the production process dimension; The temperature, relative humidity, light intensity, oxygen concentration, vibration frequency of the storage environment dimension; The content change under different storage time, the amount of related substances generated, the disintegration time limit, the dissolution, the hardness, the friability of the historical stability dimension.

[0035] It needs to be explained that these parameters construct a complete data chain from the essence of the raw material, the production process control, the storage environment influence and the quality decay law of four core dimensions, which can fully reflect the whole life cycle characteristics of the antitumor tablet from production to storage, provide multi-perspective input data for the stability prediction model, and ensure that the model can capture the key factors affecting stability.

[0036] It needs to be further pointed out that the data fusion processing module will align the data by batch when extracting parameters, ensure that the same batch of raw materials, processes, environments and stability data form a one-to-one corresponding feature matrix, and at the same time, the data is standardized and converted, such as normalizing the pressure, temperature and other parameters to the [0, 1] interval, and eliminating the data points deviating from the normal range through the outlier detection algorithm, to improve the accuracy of subsequent model construction.

[0037] S3, the stability model construction module constructs a tablet stability prediction benchmark model according to the extracted multi-dimensional characteristic parameters, marks the key influence factors, the stability critical threshold and the theoretical decay curve. In this embodiment, it needs to be specifically pointed out that the calculation formula of the key influence factor is: , is the weight of the ith parameter, is the measured value of the ith parameter in the kth batch, is the average value of the ith parameter in multiple batches, is the stability comprehensive score of the kth batch, which is calculated based on the indexes in the historical stability detection data, m is the total number of parameters, and n is the number of batches. The calculation formula of the stability critical threshold is: , is the critical threshold of the jth stability index, is the qualified standard value of the jth index, is the standard deviation of the jth index in the multi-batch historical data; The calculation formula of the theoretical decay curve is: , S0 is the initial stability index value, Wi is the key factor influence weight, Ki is the attenuation number of the i-th key factor, which is obtained by fitting historical data, ai is the time index of the i-th key factor, reflecting the attenuation acceleration, and t is the storage time, with a unit of days.

[0038] It needs to be explained that the key influence factor weight calculation formula determines the weight by quantifying the correlation between parameter fluctuation and stability score. The greater the parameter fluctuation and the worse the corresponding batch stability, the higher the weight. The numerator measures the contribution of a single parameter to stability, and the denominator normalizes to ensure that the sum of the weights is 1. For example, if the moisture content of a batch of raw materials is much higher than the average value and the stability score is low, the moisture content weight will increase. The stability critical threshold calculation formula is based on the statistical principle of normal distribution, and adds a statistical margin of 95% confidence interval based on the qualified standard value, which meets the regulations and covers production fluctuations. For example, when the related substance standard is 0.1% and the standard deviation is 0.02%, the threshold is 0.14%, which provides a fault tolerance space for production. The theoretical attenuation curve formula is based on an exponential decay model of multiple factors acting together, which integrates key factor weight, attenuation coefficient, and time index. The attenuation coefficient of temperature reflects the influence of temperature on the decay rate, and the time index reflects the acceleration trend of factors such as humidity.

[0039] It needs to be further explained that Yk in the calculation of key influence factor weight is synthesized by AHP, and the weights of content change, related substance generation, and disintegration time limit are 0.4, 0.3, and 0.2 respectively. The absolute value is calculated to reduce the influence of abnormal values. The stability critical threshold is first transformed by Box-Cox for skewed distribution indicators, and a warning threshold of 90% confidence interval is added for key indicators to intervene in advance. The theoretical attenuation curve only takes the first 5 key factors to balance complexity and accuracy, and different dosage forms need to re-fit parameters. In practical application, the curve will also be dynamically adjusted according to real-time environmental parameters to improve prediction accuracy.

[0040] S4, the real-time monitoring data acquisition module acquires the current storage environment data and real-time quality detection data of the tablet to be predicted, and transmits them to the data fusion processing module to calculate the real-time attenuation index of each stability indicator under the current conditions.

[0041] In this embodiment, it needs to be specifically explained that the real-time attenuation index of each stability indicator includes: content attenuation index, related substance growth attenuation index, and disintegration time limit attenuation index. Specifically, the calculation formula of the content attenuation index is: , Ac is the real-time attenuation index of the content, C real is the current content detection value, C0 is the initial content value, T real is the current storage temperature, T0 is the theoretical storage temperature, i.e. the standard temperature in the benchmark model, H real represents the current storage humidity, H0 represents the theoretical storage humidity, and W1 and W4 are the key influence factor weights in the benchmark model.

[0042] The formula for calculating the attenuation index of the related substance growth is: AI is the real-time attenuation index of the related substance growth, I is the current related substance generation, TI is the critical threshold of the related substance stability, t is the current storage time, T is the current storage temperature, T0 is the theoretical storage temperature, K1 is the temperature attenuation coefficient, M is the current raw material moisture content, M0 is the initial moisture content, and W2 is the key factor influence weight of the raw material moisture content.

[0043] The formula for calculating the attenuation index of the disintegration time limit is: AD is the real-time attenuation index of the disintegration time limit, D is the current disintegration time limit, D0 is the initial disintegration time limit, P is the current tabletting pressure corresponding residual influence value, P0 is the reference tabletting pressure, L is the current light intensity, L0 is the standard light intensity, W3 is the key factor influence weight of the tabletting pressure, and W5 is the influence weight of the light intensity. It needs to be explained that the content attenuation index calculation formula integrates the influence of actual content loss and environmental deviation, reflects the basic attenuation through the ratio of the current content to the initial value, and superimposes the influence of temperature and humidity deviation from the theoretical value, so that the index can fully reflect the actual situation of content attenuation. For example, when the current content is 95% and the temperature is 5°C higher than the theoretical value, the index will increase accordingly, intuitively presenting the degree of attenuation. The related substance growth attenuation index not only considers the proportion of the current related substance generation and the critical threshold, but also takes into account the influence of temperature attenuation coefficient, storage time and moisture content deviation, and can dynamically reflect the growth trend of the related substance with the environment and time. When the storage time is extended or the temperature is increased, the index will increase significantly, providing early warning of risks. The disintegration time limit attenuation index combines the actual changes of the disintegration time limit, the residual influence of the tabletting pressure and the effect of the light intensity, and reflects the attenuation of the disintegration performance in multiple dimensions. For example, when the current disintegration time limit is extended or the light is too strong, the index increases, reflecting the decline in disintegration performance.

[0044] It needs to be further explained that the key influence factor weights W1 to W5 involved in the calculation of each index come from the key influence factor weights marked in the stability prediction benchmark model, ensuring consistency with the benchmark model; the real-time collection frequency of environmental parameters such as temperature and humidity is 1 time / 5 minutes, and the mean value is taken to reduce the interference of instantaneous fluctuations on the index; real-time quality detection data such as C and I are determined by laboratory precision instruments, and the data is uploaded after being verified by the LIMS system to ensure the accuracy of the index calculation; for the tabletting pressure residual influence value P, the tabletting parameters of this batch in the production process database are associated, and the residual influence after attenuation is calculated based on the storage time, so that the index is more consistent with the actual state of the tablets.

[0045] S5, the stability deviation calculation module compares the real-time attenuation index with the theoretical attenuation benchmark in the stability prediction benchmark model, and calculates the deviation threshold range of each index. In this embodiment, it needs to be specifically pointed out that the content attenuation index deviation threshold range calculation formula is: , respectively, the lower limit and the upper limit of the content attenuation index deviation threshold, Ac real is the content real-time attenuation index, is the standard deviation of the content attenuation index in the theoretical attenuation benchmark, which is calculated based on multiple batches of historical theoretical data. The related substance growth index deviation threshold range calculation formula is: , respectively, the lower limit and the upper limit of the related substance growth index deviation threshold, AI real is the related substance growth real-time attenuation index, is the standard deviation of the related substance growth index in the theoretical attenuation benchmark. The disintegration time limit index deviation threshold range calculation formula is: , respectively, the lower limit and the upper limit of the disintegration time limit index deviation threshold, AD real is the disintegration time limit implementation attenuation index, is the standard deviation of the disintegration time limit attenuation index in the theoretical attenuation benchmark, which is calculated based on multiple batches of historical theoretical data. It needs to be explained that in the calculation formula of the disintegration time limit index deviation threshold range, the coefficient of 1.3 x sD theory is set between 1.2 times of the content attenuation and 1.5 times of the related substance growth, which not only considers the important influence of the disintegration time limit on the effectiveness of the tablet, but also sets a reasonable theoretical fluctuation coefficient combined with its fluctuation characteristics; The fixed error correction value of 0.08 is determined based on the system error analysis of the disintegration time limit detection in the historical data, so as to ensure that the threshold range can cover the normal deviation in the detection process; For example, if AD real = 0.387, sD theory = 0.04, , that is, the deviation threshold range of the disintegration time limit index is [0.255, 0.519], which not only includes the reasonable fluctuation of the real-time index, but also can effectively identify the abnormal deviation beyond the normal range. It needs to be further explained that the calculation of sD theory uses at least 30 batches of historical theoretical attenuation data, and the standard deviation is calculated after removing the abnormal values outside 3 times the standard deviation, so as to reduce the interference of extreme data on the theoretical fluctuation range; At the same time, sD theory and the fixed error correction value will be recalibrated every 5 batches of historical data, so that the deviation threshold range can dynamically adapt to the long-term changes of production process and storage conditions, and ensure the accuracy and applicability of the disintegration time limit deviation judgment.

[0046] S6, the stability deviation index analysis module comprehensively analyzes the stability deviation index based on the deviation threshold range of multiple indexes, and measures the deviation degree of the current tablet state and the theoretical model. The stability deviation index calculation formula is: SDI is the stability deviation index, n is the number of indicators involved in the calculation, Wj is the weight of the j-th indicator, Ajreal is the real-time decay index of the j-th indicator, and Ajtheoretical is the theoretical decay benchmark of the j-th indicator. Let be the width of the deviation threshold range for the j-th indicator. An example calculation shows that... It needs to be explained that the stability deviation index calculation formula integrates the deviation of various indicators into a single index through standardization, where the numerator... The denominator measures the absolute deviation between the real-time decay index and the theoretical decay benchmark. The deviation is standardized by defining the range of deviation thresholds to eliminate the influence of differences in the magnitude of different indicators. Then, a weighted sum is performed using the indicator weights Wj, allowing the index to comprehensively reflect the overall situation of multi-dimensional deviations. For example, the weight of the content decay indicator is 0.4, and the weight of the related substances growth indicator is 0.6, reflecting the higher priority of related substances in influencing stability. When the related substances deviation is large, it will dominate the change in the SDI value, which is more in line with the logic of actual quality risk assessment. It needs further explanation that the number of indicators n involved in the calculation is dynamically adjusted according to the tablet type. For high-risk formulations, all three core indicators (content, related substances, and disintegration time) are included, while for ordinary formulations, this can be simplified to two key indicators. The indicator weights Wj are determined using the Delphi method combined with expert scoring. The weights of related substances, content, and disintegration time are usually set to 0.6, 0.3, and 0.1, respectively, to strengthen the influence of safety indicators. During the calculation process, if the real-time decay index of a certain indicator exceeds the deviation threshold range, i.e., > If the deviation percentage of this indicator is 1.0, it avoids excessive dilution of extreme deviations. Furthermore, the SDI threshold classification—SDI < 0.2 for slight deviation, SDI between 0.2 and 0.5 for moderate deviation, and SDI > 0.5 for significant deviation—is determined based on deviation analysis results from historical batch data. This ensures accurate mapping of the deviation between the actual quality state of the tablets and the theoretical model, providing a quantitative basis for subsequent model optimization or correction. In the S7 prediction result generation and dynamic update module, if the deviation exceeds the threshold based on the stability deviation index, the model's dynamic optimization algorithm is triggered to regenerate the stability prediction curve; if the deviation is small, local corrections are made based on the baseline model.

[0047] In this embodiment, it should be specifically noted that the deviation threshold is set to 0.2. This threshold is determined based on statistical analysis of multiple batches of historical data. The stability deviation index (SDI) between 0.2 and 0.5 indicates moderate deviation, and SDI > 0.5 indicates significant deviation. When the stability index SDI ≥ 0.2, the model dynamic optimization algorithm is triggered; when SDI < 0.2, it is determined to be a slight deviation, and local correction is performed. The model dynamic optimization algorithm adopts an improved particle swarm optimization algorithm combined with a Bayesian network. By introducing adaptive inertia weights and mutation operations, it avoids getting trapped in local optima and re-optimizes the weights of key influencing factors, the critical stability threshold, and the theoretical decay curve parameters. The local correction is aimed at the real-time fluctuations of storage environment parameters. While retaining the core structure of the baseline model, the prediction curve is fine-tuned by adjusting the local slope of the decay curve.

[0048] It should be explained that setting the deviation threshold to 0.2 is a balance point determined after comprehensively considering the model prediction error and actual production fluctuations. Historical data shows that when SDI ≥ 0.2, the average deviation between the model prediction and the actual detection value exceeds 5%, which has affected the accuracy of the expiration date prediction. When SDI < 0.2, the deviation mainly stems from short-term fluctuations in environmental parameters, and the difference between prediction and reality can be effectively bridged through local correction. In the model's dynamic optimization algorithm, the particle swarm optimization algorithm is responsible for globally searching for the optimal parameter combination, while the Bayesian network updates the parameter distribution based on prior knowledge and real-time data. The combination of the two can ensure optimization efficiency while avoiding overfitting. For example, when the growth of related substances in a batch of tablets is abnormal, the algorithm will automatically increase the weight of related factors such as temperature and moisture, and adjust the critical threshold of related substances to make the newly generated prediction curve closer to the actual degradation trend.

[0049] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0050] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting the stability of antitumor drug tablets based on multi-source data fusion, characterized in that, The method includes a multi-source data acquisition module, a tablet stability database, a data fusion processing module, a stability model construction module, a real-time monitoring data acquisition module, a stability deviation calculation module, a stability deviation index analysis module, and a prediction result generation and dynamic update module. The specific steps of the method are as follows: S1. The multi-source data acquisition module collects raw data from multiple batches of antitumor drug tablets and uploads the raw data to the tablet stability database. S2. The data fusion processing module extracts multi-dimensional feature parameters of tablets from the tablet stability database in batches and transmits them to the stability model construction module. The multi-dimensional characteristic parameters specifically refer to the content of active pharmaceutical ingredient in the raw material, purity of excipients, particle size distribution, crystal structure, moisture content, and density of the raw material characteristics. Production process dimensions include granulation stirring speed, granulation time, tableting pressure, tableting speed, drying temperature, drying time, coating thickness, and coating curing time; storage environment dimensions include temperature, relative humidity, light intensity, oxygen concentration, and vibration frequency; historical stability dimensions include content changes, related substance formation, disintegration time, dissolution rate, hardness, and friability under different storage durations. S3. The stability model construction module constructs a tablet stability prediction benchmark model based on the extracted multi-dimensional feature parameters, and marks key influencing factors, stability critical thresholds and theoretical decay curves. The formula for calculating the key impact factor is as follows: , The weight of the i-th parameter, Let be the measured value of the i-th parameter in the k-th batch. The average value of the i-th parameter across multiple batches. The stability score for the kth batch is given by m, where m is the total number of parameters and n is the number of batches. The formula for calculating the stability critical threshold is: , Let j be the critical threshold of the stability index. Let j be the pass / fail value for the j-th indicator. The standard deviation of the j-th indicator in multiple batches of historical data; The formula for calculating the theoretical attenuation curve is: , S0 is the theoretical value of the stability index at storage time t, Wi is the influence weight of the key factor, Ki is the decay number of the i-th key factor, ai is the time exponent of the i-th key factor, reflecting the decay acceleration, and t is the storage time in days. S4. The real-time monitoring data acquisition module collects the current storage environment data and real-time quality detection data of the batch of tablets to be predicted in real time, and transmits them to the data fusion processing module to calculate the real-time decay index of various stability indicators under the current conditions. The real-time decay indices of the stability indicators include: content decay index, related substance growth decay index, and disintegration time decay index. The formula for calculating the content decay index is: Ac is the real-time decay index of content, C is the current content detection value, C0 is the initial content value, T is the current storage temperature, T0 is the theoretical storage temperature, H is the current storage humidity, H0 is the theoretical storage humidity, and W1 and W4 are the weights of key influencing factors in the benchmark model. The formula for calculating the growth and decay index of the relevant substance is as follows: AI is the real-time decay index of related substance growth, I is the current amount of related substance generated, TI is the critical threshold of related substance stability, t is the current storage time, T is the current storage temperature, T0 is the theoretical storage temperature, K1 is the temperature decay coefficient, M is the current measured value of raw material moisture content, M0 is the initial moisture content, and W2 is the key factor influence weight of raw material moisture content. The formula for calculating the disintegration time decay index is as follows: AD is the real-time decay index of disintegration time limit, D is the measured value of the current disintegration time limit, D0 is the initial disintegration time limit, P is the residual influence value corresponding to the current tableting pressure, P0 is the reference tableting pressure, L is the current light intensity, L0 is the standard light intensity, W3 is the key factor influence weight of tableting pressure, and W5 is the influence weight of light intensity. S5, the stability deviation calculation module compares the real-time decay index with the theoretical decay benchmark in the stability prediction benchmark model and calculates the deviation threshold range of each index. S6. The stability deviation index analysis module analyzes the deviation threshold range of multiple indicators to obtain the stability deviation index, which measures the degree of deviation between the current tablet state and the theoretical model. The formula for calculating the stability deviation index is as follows: SDI is the stability deviation index, n is the number of indicators involved in the calculation, Wj is the weight of the j-th indicator, Ajreal is the real-time decay index of the j-th indicator, and Ajtheoretical is the theoretical decay benchmark of the j-th indicator. is the interval width of the deviation threshold range for the j-th indicator; S7. The prediction result generation and dynamic update module, based on the stability deviation index, if the deviation exceeds the threshold, triggers the model dynamic optimization algorithm to regenerate the stability prediction curve; if the deviation is small, it performs local correction based on the baseline model.

Images