An oral liquid medicine production whole-process quality tracing method and system

Abstract

This invention relates to the field of pharmaceutical quality traceability technology, and particularly to a method and system for full-process quality traceability of oral liquid pharmaceutical production. The method includes: acquiring online sensor data to establish a parameter time-series database; collecting temperature, pressure, stirring motor frequency, and nitrogen charging delay time; calculating the integral and percentage of these data and combining them into a process deviation feature vector; extracting historical batch content decay slopes to construct a sample set; establishing a correlation model between the feature vector and the decay slope; inputting the current batch feature vector into the model to calculate the predicted stability acceleration decay slope; determining a judgment threshold based on historical distribution and generating batch risk markers; generating differentiated traceability record packages based on the markers; adding release decision support suggestions; and incorporating regular batch data into the historical sample database. This invention advances the data granularity to the time-series deviation integral, achieving implicit oxidation risk prediction before batch release.

Description

Technical Field

[0001] This invention relates to the field of pharmaceutical quality traceability technology, and in particular to a method and system for tracing the quality of the entire production process of oral liquid pharmaceuticals. Background Technology

[0002] Oral liquid drugs belong to a category of liquid preparations with relatively complex manufacturing processes, especially for low-dose, high-activity ingredient formulations. Their production process involves multiple key steps, including cleanroom environmental control, inert gas protection preparation, multi-stage filtration, and precision filling. As drug regulations become increasingly stringent, pharmaceutical companies are placing higher demands on the traceability of data throughout the production process. A full-process quality traceability system has become a fundamental support for the compliant production of oral liquid drugs.

[0003] To address the aforementioned traceability needs, the industry's existing technological approach primarily employs electronic batch record systems to archive operational data for each process. These systems operate on a process node basis, requiring operators to input key parameters such as preparation temperature, stirring time, filling volume, and filter cartridge batch number after completing their tasks. All records are then aggregated and stored at the batch level, forming a traceability archive available for regulatory review. Some companies have also introduced online sensors to collect real-time data on parameters such as the temperature and pressure of the mixing tank and automatically write this data into the batch records, reducing manual data entry errors.

[0004] Chinese patent application CN114742445A discloses an information-enabled drug traceability method and system. This system uses a central control module to compare actual measured parameters at each stage of the entire supply chain with preset reasonable parameter standards. When actual parameters exceed or fall below the preset standards, the system is deemed unqualified and marked on the traceability path, thus achieving quality traceability from raw materials to distribution. This solution represents the current traceability technology route centered on parameter specification limit comparison.

[0005] However, for products with extremely low dosage and high oxidation sensitivity such as calcitriol oral solution, the existing traceability system has a long-neglected defect in extreme quality deviation scenarios: the system uses whether a parameter exceeds the upper and lower specification limits as the only basis for quality judgment, and can only identify single-point explicit out-of-specification events, but cannot capture the cumulative amount of time-series deviation of parameters within the specification limits. Taking the compounding process as an example, the oxidation rate of calcitriol raw material increases significantly when the temperature exceeds 25°C. If the temperature is maintained near 24.5°C for a long time during the compounding cycle, the cumulative oxidation exposure far exceeds that of batches operated at 20°C, but all existing systems judge such batches as qualified. Similarly, the time sequence delay of nitrogen filling relative to the start of stirring will cause the liquid medicine to be briefly exposed to oxygen under atmospheric pressure, and this time window also cannot be captured by the single-point pressure parameter. The above-mentioned hidden process deviations cannot be detected during batch release inspection because the time effect has not been fully manifested, and they are only exposed in the form of accelerated content attenuation during the stability retention sample investigation. By this time, the batches have already been distributed, and traceability intervention has lost its practical significance. Summary of the Invention

[0006] To solve the technical problem in the prior art that existing electronic batch record systems cannot capture the cumulative amount of time-series deviations within specification limits, resulting in blind spots in quality traceability for high oxidation-sensitive oral solution varieties, the present invention provides a full-process quality traceability method and system for oral solution pharmaceutical production.

[0007] In a first aspect, the present invention provides a full-process quality traceability method for oral solution pharmaceutical production, which adopts the following technical solution:

[0008] S1, acquiring online sensing data of each process in the full production process of oral solution pharmaceuticals, establishing a parameter time series database, collecting temperature series, pressure series, stirring motor frequency series and nitrogen filling delay time in the compounding process, respectively calculating overtemperature deviation integral, negative pressure deviation integral and proportion of stirring frequency deviation duration, and combining the foregoing with the nitrogen filling delay time into a process deviation feature vector; S2, extracting content attenuation slopes from a stability retention sample database of historical batches, constructing a sample set in combination with the process deviation feature vectors corresponding to the historical batches, and establishing an association model between the process deviation feature vectors and the content attenuation slopes; S3, inputting the process deviation feature vector of the current batch into the association model, calculating a predicted stability accelerated attenuation slope, determining a judgment threshold based on the distribution of accelerated attenuation slopes of historical batches, and generating a batch risk mark according to the positional relationship between the predicted stability accelerated attenuation slope and the judgment threshold; S4, generating differentiated traceability record packages based on batch risk marks, adding a release decision support suggestion to the traceability record package for batches marked as having hidden deviation risks, and incorporating the process deviation feature vectors corresponding to batches marked as conventional batches into a historical sample library to update the judgment threshold.

[0009] By advancing the data granularity of oral liquid production traceability from batch snapshots to time-series deviation integrals, a correlation model is constructed between the process deviation feature vector and the historical stability decay slope. This model can identify batches with process parameters within acceptable limits but high cumulative exposure levels and hidden oxidation risks before batch release.

[0010] Preferably, the integral of over-temperature deviation, the integral of negative pressure deviation, and the proportion of stirring frequency deviation duration are calculated separately, including: extracting the positive deviation portion exceeding the upper limit of temperature control in the temperature sequence, integrating the positive deviation portion over time to obtain the integral of over-temperature deviation; extracting the negative deviation portion below the micro-positive pressure reference in the pressure sequence, integrating the negative deviation portion over time to obtain the integral of negative pressure deviation; counting the number of sampling points deviating from the rated operating range in the stirring motor frequency sequence, calculating the ratio of the number of sampling points to the total number of sampling points, and finally obtaining the proportion of stirring frequency deviation duration.

[0011] By calculating the integral of over-temperature deviation, the integral of negative pressure deviation, and the proportion of stirring frequency deviation duration, the Boolean description of whether parameters exceed limits in traditional batch records is upgraded to a continuous cumulative quantity, so that implicit process deviations can be clearly distinguished at the numerical level.

[0012] Preferably, the correlation model between the process deviation feature vector and the content decay slope is established, including: fitting the sample set with a multiple linear regression method to obtain the contribution coefficient vector and intercept term of each feature component to the content decay slope; performing a statistical significance test on each component in the contribution coefficient vector, and setting the coefficients corresponding to feature components with insignificant test results to zero; when the number of available historical batches is less than a preset threshold, finding the nearest neighbor batch with the feature vector closest to the current batch in the historical batches, using the average decay slope of the nearest neighbor batches as the risk reference slope, and finally obtaining the correlation model.

[0013] By establishing a correlation model between the process deviation feature vector and the content decay slope, a quantitative mapping relationship is established between the two types of heterogeneous data stored separately, so that the potential impact of process deviation on the long-term stability of the product can be numerically predicted before batch release.

[0014] Preferably, the determination threshold is based on the distribution of the accelerated decay slope of historical batches, including: extracting the absolute value distribution of the slope of the accelerated decay slope of historical batches from the stability assessment conclusion of the qualified group; calculating the first quantile of the absolute value distribution of the slope as the upper threshold for risk determination; calculating the second quantile of the absolute value distribution of the slope as the lower threshold for boundary warning, and finally obtaining the determination threshold.

[0015] By extracting the absolute value of the slope of the qualified group in the accelerated decay slope distribution of historical batches, and calculating the first quantile as the upper threshold for risk judgment and the second quantile as the lower threshold for boundary warning, the adaptive statistical determination of the judgment threshold is realized. The upper threshold is used to identify batches with hidden oxidation risks, and the lower threshold is used to delineate the boundary between regular batches and batches requiring attention, forming a two-level warning mechanism. This provides an objective statistical basis for subsequent batch risk marking that is continuously optimized with the accumulation of historical data.

[0016] Preferably, the batch risk marker is generated based on the positional relationship between the predicted stability acceleration decay slope and the judgment threshold, including: generating a latent deviation risk batch marker when the absolute value of the predicted stability acceleration decay slope is greater than the upper risk judgment threshold; generating a boundary concern batch marker when the absolute value of the predicted stability acceleration decay slope is between the lower boundary warning threshold and the upper risk judgment threshold; and generating a regular batch marker when the absolute value of the predicted stability acceleration decay slope is less than the lower boundary warning threshold, thus finally obtaining the batch risk marker.

[0017] By generating batch risk markers based on the positional relationship between the predicted stability acceleration decay slope and the judgment threshold, an additional quantitative predictive signal for future stability is provided, enabling batches with latent oxidation risks to be identified and intervened before entering the circulation process.

[0018] Preferably, generating batch risk markers based on the positional relationship between the predicted stability acceleration decay slope and the judgment threshold further includes: calculating the contribution percentage of each component of the process deviation feature vector to the predicted stability acceleration decay slope; identifying the deviation component with the highest contribution percentage as the main risk source; and attaching the main risk source to the batch risk marker, ultimately obtaining a batch risk marker containing feature vector decomposition descriptions.

[0019] Preferably, the differentiated traceability record package is generated based on batch risk marking, including: generating a first-level process-level operation compliance record containing the comparison results of process parameter specification limits; generating a second-level process deviation analysis record containing process deviation feature vectors, predicted stability acceleration decay slope and threshold comparison results; generating a third-level risk source marking based on the operation records and operator information associated with the main risk sources, and finally obtaining the differentiated traceability record package.

[0020] Preferably, release decision support recommendations are added to the traceability record package, including: generating recommendations for relevant testing items to supplement the assessment of actual antioxidant consumption; generating recommendations to include the corresponding batch in the enhanced stability retention plan; and pushing the relevant testing item recommendations and the enhanced stability retention plan recommendations through the data interface to finally obtain release decision support recommendations.

[0021] Preferably, the process deviation feature vectors corresponding to the regular batches are included in the historical sample library to update the judgment threshold. This includes: automatically storing the process deviation feature vectors of the regular batches and the stability retention data continuously collected after release into the historical sample library; re-statistically analyzing the distribution of qualified groups based on the expanded historical sample library; recalculating the quantiles based on the updated distribution of qualified groups, and finally obtaining the updated judgment threshold.

[0022] By automatically storing the process deviation feature vectors of regular batches and the stability retention data continuously collected after release into the historical sample library, the training dataset of the correlation model is continuously expanded as production batches accumulate naturally. On this basis, the distribution of qualified groups is re-statistically analyzed and the quantile threshold is updated, achieving the technical effect that the judgment threshold gradually converges to a more accurate level as historical data becomes richer.

[0023] Secondly, this invention provides a quality traceability system for the entire production process of oral liquid drugs, employing the following technical solution: A quality traceability system for the entire production process of oral liquid drugs includes a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the system implements the quality traceability method for the entire production process of oral liquid drugs as described above.

[0024] The present invention has the following technical effects: This invention enables traceability records to include quantifiable, in-depth process quality information for the first time, effectively extending the predictive capabilities of batch traceability records. It provides more comprehensive data support for quality-authorized release decisions and reduces the possibility of batches with latent oxidation risks entering the distribution process. As historical batch data continues to accumulate, the judgment capability of the correlation model will improve accordingly. The traceability system has an inherent ability to improve naturally with production data, fundamentally filling the quality traceability blind spots of existing systems for extremely low-dose, highly oxidation-sensitive oral liquid products. Attached Figure Description

[0025] Figure 1 This is a flowchart of a method for quality traceability throughout the entire production process of oral liquid drugs according to the present invention.

[0026] Figure 2 This is a scatter plot comparing the accuracy of existing methods in predicting the slope of content decay.

[0027] Figure 3 This is a scatter plot comparing the accuracy of the method of the present invention in predicting the content decay slope. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0029] This invention discloses a method for quality traceability throughout the entire production process of oral liquid drugs, referring to... Figure 1 This includes the following steps: S1. Collect data and combine it into a process deviation feature vector.

[0030] In one optional embodiment, online sensor data from each step of the entire oral liquid drug production process is acquired, and a parameter time series database is established using timestamps as indexes. For the preparation step, temperature sequences within the preparation tank are collected at preset sampling intervals. Internal pressure sequence and the frequency sequence of the stirring motor For example, the sampling interval can be set to no more than 30 seconds; at the same time, the opening time of the nitrogen charging valve is recorded. With the start time of stirring Calculate the nitrogen charging delay time. . The oxygen exposure time window of the drug solution under normal pressure is used to characterize the active ingredient. Because calcitriol is extremely sensitive to oxidation, even tens of seconds of exposure at normal pressure can leave an irreversible oxidation starting point at the active ingredient level. Therefore, this parameter is recorded as an independent dimension and is not combined with the pressure sequence. There are various mature options for the deployment of online sensors and communication protocols in this field, and those skilled in the art can flexibly select the appropriate configuration based on the actual production line conditions. This solution does not limit the hardware implementation.

[0031] Furthermore, for the temperature series, the integral of the overtemperature deviation is calculated. Since calcitriol does not exhibit a significant oxidation-accelerating effect at low temperatures, it is only used when the temperature exceeds the upper limit of the temperature control during the feeding stage of the preparation process specified in the process specification. Only when the oxidative damage to the active ingredient is present does it have substantial significance. Therefore, the integral only accumulates the positive deviation portion. The positive deviation portion exceeding the upper limit of temperature control in the temperature sequence is extracted, and the positive deviation portion is integrated over time to obtain the over-temperature deviation integral. The specific calculation is as follows:

[0032] in, For the first Measured temperature values ​​at each sampling time This corresponds to the upper limit of temperature control specified in the process procedure for that stage. The sampling interval is... This represents the total number of sampling points in this process, in °C·min. The specific values ​​are verified during each company's process development phase and then formalized in the approved process specifications. This solution directly references the values ​​specified in the process specifications and does not specify the exact values ​​here. Over-temperature deviation integral. The higher the value, the longer the medicine solution is kept above the upper limit of the safe temperature during the preparation process, the deeper the degree of heat exposure, and the more significant the potential amount of oxidation accumulation.

[0033] Next, for the pressure sequence, analogous to the calculation logic of temperature deviation integral, the pressure sequence below the micro-positive pressure reference is extracted. The negative deviation portion is then integrated over time to obtain the integral of the negative deviation quantity, which represents the pressure negative deviation during the preparation process where the pressure inside the tank is lower than the micro-positive pressure reference. This is used to characterize the cumulative oxygen exposure of the drug solution. The value is the minimum maintaining pressure lower limit of the preparation tank specified in the process procedure. It represents the lowest overpressure value inside the tank sufficient to prevent external air from seeping in; the specific value is determined by each company's process documentation. Statistical analysis is performed on the stirring motor frequency sequence. The number of sampling points that deviate from the rated operating range of the stirring frequency specified in the process specification is calculated, and the ratio of this number of sampling points to the total number of sampling points is divided by the total number of sampling points in this process. The final result is the percentage of time the stirring frequency deviates from the rated range. , which is a dimensionless ratio. The higher the value, the higher the proportion of time the stirring speed operates outside the process window, and the more serious the deviation of the mixing uniformity and dissolved oxygen state from the process design intent.

[0034] Furthermore, the integral of the over-temperature deviation Integral quantity of negative pressure deviation Nitrogen delay time Percentage of stirring frequency deviations over time Combined into the process deviation feature vector of the current batch :

[0035] The four components independently quantify the latent oxidation risk of the same batch from four dimensions: heat exposure, atmosphere protection failure, initial oxygen exposure, and process deviation. Together, they form the input data structure for subsequent correlation modeling. The absence of any component will result in the loss of risk dimensions.

[0036] In this way, by sampling the time-series parameters of the entire preparation process point by point and calculating the deviation integral in different dimensions, the Boolean description of whether the parameters exceed the limits in the traditional batch record is upgraded to a continuous cumulative quantity. This allows implicit process deviations such as long-term high temperature but not exceeding the standard and delayed nitrogen charging to be clearly distinguished at the numerical level, providing physically meaningful and information-rich feature inputs for subsequent correlation modeling.

[0037] S2. Extract the historical content decay slope and establish a correlation model.

[0038] In an optional embodiment, the content decay slope of each batch under accelerated and long-term testing conditions is extracted from a historical batch stability database. Specifically, with calcitriol content as the dependent variable, linear regression is performed on the measured content data of each batch at 0, 1, 2, 3, 6, 9, and 12 months of observation under accelerated testing conditions (40℃ / 75%RH), and the resulting slope is denoted as the accelerated decay slope. Linear regression was performed on the content data of the same time point sequence under long-term experimental conditions (25℃ / 60%RH), and the resulting slope was recorded as the long-term decay slope. All figures are in percentages per month. and The larger the absolute value, the more severe the potential oxidative damage accumulated during the production process, and the more likely the stability test result is to be unacceptable. Simultaneously, feature vectors of process deviations formed during the production of each historical batch are extracted. A sample set for regression modeling is constructed by combining the process deviation feature vectors corresponding to historical batches. .

[0039] Furthermore, based on the extracted sample set, the existing least squares multiple linear regression method is used to fit the sample set, establishing a correlation model between the process deviation feature vector and the accelerated decay slope, and fitting the contribution coefficient vector of each feature component to the content decay slope. and intercept term .in correspond , correspond , correspond , correspond Each coefficient reflects the actual impact of different types of process deviations on stability degradation; their values ​​are obtained by fitting historical data, not by arbitrary specification. Intercept term. The baseline decay rate of the variety itself is determined by regression analysis of historical samples when the process parameters are completely unbiased.

[0040] After the model is built, the contribution coefficient vector needs to be... Each component in the model was subjected to a statistical significance test. Since the actual impact of different types of process deviations on the stability of calcitriol may vary, the explanatory power of some deviation components on the decay slope in historical data may not reach a statistical significance level. For feature components with insignificant test results, their corresponding coefficients were set to zero and removed from the prediction formula to prevent invalid features from interfering with the model's predictive stability.

[0041] Next, to address the issue of insufficient historical sample batches, an automatic degradation mechanism is implemented. When the number of available historical batches falls below a preset threshold determined based on the minimum statistical requirements of the regression model for sample size, the regression model risks overfitting due to sample sparsity. For example, this threshold can be five times the feature vector dimension, i.e., 20 batches. Upon triggering degradation, the system automatically switches to a nearest neighbor retrieval mode based on Euclidean distance: searching for features in historical batches that match the current batch's feature vector. Euclidean distance nearest A neighboring batch, The value is taken as the total number of available samples in the historical database at that time. Round down to the square root, that is And not exceeding a preset upper limit value, for example, this upper limit value can be 5; The average accelerated decay slope of the nearest neighboring batches is used as the risk reference slope for the current batch, directly applied to subsequent risk assessment, ultimately yielding the correlation model. For example, if the total number of available samples in the historical database at that time... If there are 9 batches, then That is, the three historical batches with the closest Euclidean distance are retrieved to participate in the mean calculation.

[0042] In this way, by correlating and fitting the stability sample data of historical batches with the deviation feature vectors of the corresponding production batches, a quantitative mapping relationship is established for the first time between the two types of heterogeneous data that were originally stored separately in the stability database and the batch production records. This allows the potential impact of process deviations on the long-term stability of the product to be numerically predicted before the batch is released.

[0043] S3. Calculate the attenuation slope and generate batch risk markers.

[0044] In an optional embodiment, before the current batch production ends and before batch release inspection begins, the calculated process deviation feature vector of the current batch is... Input the established correlation model and calculate the predicted stability acceleration decay slope for this batch. The complete form of the prediction is:

[0045] in The coefficient vector obtained by fitting the correlation modeling step, The intercept term is obtained from the synchronous fitting, and the meanings of each variable are consistent with those in the data acquisition steps.

[0046] Furthermore, after obtaining The system then compares this with the distribution of accelerated decay slopes from historical batches, automatically determining two judgment thresholds statistically. The system extracts the absolute value distribution of the slopes from the historical batch accelerated decay slope distributions, focusing on the stability assessment results for the qualified group. The first quantile of this absolute value distribution is then calculated and used as the upper threshold for risk assessment. For example, the first quantile can be the 95th quantile; the second quantile of the absolute value distribution of the slope is calculated and used as the threshold for boundary warning. For example, the second quantile can be the 50th quantile, ultimately yielding the decision threshold. Both thresholds are automatically determined by historical data statistics and are dynamically updated as the historical batch database continues to expand.

[0047] Next, based on the positional relationship between the predicted stability acceleration decay slope and the judgment threshold, a batch risk label is generated. The system assigns three types of labels to the current batch: when the absolute value of the predicted stability acceleration decay slope is... Greater than the upper threshold for risk assessment When the absolute value of the predicted stability acceleration decay slope is below the boundary warning threshold, a latent bias risk batch marker is generated; Risk assessment threshold During this period, boundary concern batch markers are generated; when the absolute value of the predicted stability acceleration decay slope is less than the boundary warning threshold... At that time, a regular batch marker is generated, and finally a batch risk marker is obtained.

[0048] All three types of markers come with complete eigenvector decomposition descriptions, and the system automatically calculates the process deviation eigenvectors. The contribution percentage of each component to the predicted stability acceleration decay slope is determined; the deviation component with the highest contribution percentage is identified as the main risk source; the main risk source is added to the batch risk label, ultimately obtaining a batch risk label containing eigenvector decomposition descriptions, providing a directional basis for subsequent process improvements. For example, if... Item in If the contribution is the highest, the system will clearly indicate in the labeling description that nitrogen charging delay is the main source of risk, prompting process personnel to focus on reviewing the timing control of this operation node.

[0049] In this way, by mapping the integral characteristics of the process deviation to the predicted attenuation slope before the batch release inspection is initiated, and by statistically comparing it with the distribution of historical qualified groups, the system can provide an additional quantitative predictive signal for future stability, on the basis that conventional release inspection can only determine the current compliance. This makes it possible to identify and intervene in batches with hidden oxidation risks before they enter the circulation process.

[0050] S4. Generate a traceability record package and update the judgment threshold.

[0051] In an optional embodiment, differentiated traceability record packages are generated based on batch risk marking. The traceability record packages organize data in three layers: a first-layer process-level operation compliance record containing comparison results of process parameter specification limits (i.e., traditional batch record content), recording whether each process parameter is within specification limits; this layer uses the same generation rules for the three types of marked batches. A second-layer process deviation analysis record containing process deviation feature vectors, predicted stability acceleration decay slopes, and threshold comparison results is generated, enabling each batch's traceability file to include quantifiable process quality depth information for the first time, rather than just Boolean values ​​indicating whether standards are exceeded. A third-layer risk source labeling is generated based on the operation records and operator information corresponding to the main risk sources, ultimately resulting in a differentiated traceability record package, forming a complete traceability chain from quality risk to personnel, machinery, materials, methods, and environment.

[0052] Furthermore, for batches marked as having latent deviation risks, release decision support recommendations are added to the traceability record package. Based on existing release inspection items, the system generates recommendations for supplementary testing to assess actual antioxidant consumption; it also generates recommendations to include the corresponding batch in the enhanced stability retention plan, adding an additional inspection point in the first month beyond the regular inspection frequency to verify or eliminate latent oxidation risks as early as possible. The specific supplementary inspection items are ultimately decided by the quality authorized person based on the current batch's formulation and risk analysis results; the system only provides data support and does not replace human decision-making. The relevant testing item recommendations and enhanced stability retention plan recommendations are pushed through the data interface to ultimately obtain release decision support recommendations.

[0053] Next, for batches marked as "regular batches," the traceability record package is archived according to standard procedures without additional release recommendations. Simultaneously, the process deviation feature vectors corresponding to these "regular batches" are incorporated into the historical sample library to update the judgment threshold. Specifically, this includes automatically storing the process deviation feature vectors of regular batches and the stability sampling data continuously collected after release into the historical sample library, continuously expanding the training dataset of the correlation model. As the number of batches entering the library accumulates, the distribution of the qualified population is recalculated based on the expanded historical sample library; quantiles are recalculated based on the updated distribution of the qualified population, ultimately obtaining the updated judgment threshold, continuously converging to a more accurate level. Because the background decay characteristics of different varieties and the degree of impact of seasonal temperature and humidity fluctuations on the production process vary, the continuous expansion of the historical sample library allows the correlation model to gradually incorporate these objective variation factors into the statistical background, thereby steadily improving the judgment accuracy with the natural accumulation of production data. For batches of boundary concern, the traceability record package sends an early warning to the quality authorized person while archiving the data. The quality authorized person then decides whether to initiate an additional review based on the actual situation of the batch. The system automatically includes the subsequent stability sampling data of the batch in the tracking and monitoring. Once the actual data shows signs of deviation, an automatic review prompt is triggered.

[0054] In this way, by mapping the batch risk marking results to a three-level differentiated traceability record package and continuously feeding back the production data of regular batches to the historical sample library, the system achieves continuous self-updating of the correlation model while completing the traceability of a single batch. This enables the traceability system to have the inherent attribute of naturally improving its judgment ability as production data accumulates, without the need for manual periodic recalibration of the judgment threshold.

[0055] Figure 2 The graph shows the scatter plot relationship between the predicted and measured values ​​of the existing method. The horizontal axis represents the measured value of the content decay slope (% / month), and the vertical axis represents the predicted value of the content decay slope (% / month). It can be observed from the graph that the scatter plot is highly dispersed relative to the ideal line, and the prediction system has obvious bias. Figure 3 The corresponding scatter plot relationship of the method of the present invention is shown. The diamond scatter plot represents the predicted value of the method of the present invention. The fitting trend line and the ideal line are significantly better than those of the existing methods, indicating that the correlation model of the present invention based on the process deviation feature vector has higher predictive consistency for the acceleration decay slope.

[0056] This invention also discloses a quality traceability system for the entire production process of oral liquid drugs, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a quality traceability method for the entire production process of oral liquid drugs according to this invention is implemented.

[0057] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for quality traceability throughout the entire production process of oral liquid drugs, characterized in that, include: S1. Obtain online sensor data for each process in the entire production process of oral liquid drugs, establish a parameter time series database, collect the temperature sequence, pressure sequence, stirring motor frequency sequence and nitrogen charging delay time of the preparation process, calculate the integral of over-temperature deviation, integral of negative pressure deviation and the proportion of stirring frequency deviation time, and combine them with nitrogen charging delay time to form a process deviation feature vector. S2. Extract the content decay slope from the stability retention database of historical batches, combine it with the process deviation feature vectors corresponding to the historical batches to construct a sample set, and establish a correlation model between the process deviation feature vectors and the content decay slope. S3. Input the process deviation feature vector of the current batch into the correlation model, calculate the predicted stability accelerated decay slope, determine the judgment threshold based on the distribution of the accelerated decay slope of historical batches, and generate batch risk markers based on the positional relationship between the predicted stability accelerated decay slope and the judgment threshold. S4. Generate differentiated traceability record packages based on batch risk marking. For batches marked as having hidden deviation risk, add release decision support suggestions to the traceability record package and include the process deviation feature vectors corresponding to batches marked as normal in the historical sample library to update the judgment threshold.

2. The method for quality traceability throughout the entire production process of oral liquid drugs according to claim 1, comprising calculating the integral of over-temperature deviation, the integral of negative pressure deviation, and the proportion of stirring frequency deviation duration, including: Extract the positive deviation portion of the temperature sequence that exceeds the upper limit of temperature control, and integrate the positive deviation portion over time to obtain the over-temperature deviation integral. Extract the negative deviation portion of the pressure sequence that is below the micro-positive pressure reference, and integrate the negative deviation portion over time to obtain the pressure negative deviation integral. The number of sampling points that deviate from the rated operating range in the frequency sequence of the stirring motor is counted, and the ratio of the number of sampling points to the total number of sampling points is calculated to obtain the percentage of the stirring frequency deviation time.

3. The method for quality traceability throughout the entire production process of oral liquid drugs according to claim 1, comprising establishing a correlation model between the process deviation feature vector and the content decay slope, including: The sample set was fitted using a multiple linear regression method to obtain the contribution coefficient vector and intercept term of each feature component to the content decay slope. Perform a statistical significance test on each component in the contribution coefficient vector, and set the coefficients of the feature components whose test results are not significant to zero. When the number of available historical batches is less than a preset threshold, the nearest neighbor batch with the feature vector of the current batch is found in the historical batches. The average decay slope of the nearest neighbor batches is used as the risk reference slope to finally obtain the correlation model.

4. The method for quality traceability of the entire production process of oral liquid drugs according to claim 1, wherein a judgment threshold is determined based on the distribution of the accelerated decay slope of historical batches, includes: The stability analysis of the distribution of accelerated decay slopes in historical batches concluded that the distribution of the absolute value of the slope of the qualified population was the result of the analysis of the distribution of the slope of the qualified population. Calculate the first quantile of the absolute value distribution of the slope, and use it as the upper threshold for risk assessment; The second quantile of the absolute value distribution of the slope is calculated and used as the threshold under the boundary warning, thus obtaining the final judgment threshold.

5. The method for quality traceability of the entire production process of oral liquid drugs according to claim 1, comprising generating batch risk markers based on the positional relationship between the predicted stability acceleration decay slope and the judgment threshold, including: When the absolute value of the predicted stability acceleration decay slope is greater than the upper threshold of risk assessment, a latent bias risk batch marker is generated. When the absolute value of the predicted stability acceleration decay slope is between the lower threshold of the boundary warning and the upper threshold of the risk judgment, a boundary attention batch marker is generated. When the absolute value of the predicted stability acceleration decay slope is less than the threshold under the boundary warning, a regular batch label is generated, and finally the batch risk label is obtained.

6. The method for quality traceability of the entire production process of oral liquid drugs according to claim 1, which generates batch risk markers based on the positional relationship between the predicted stability acceleration decay slope and the judgment threshold, further includes: The contribution percentage of each component of the process deviation eigenvector to the predicted stability acceleration decay slope is calculated. Identify the deviation component with the highest contribution as the main source of risk; The main sources of risk are appended to the batch risk label, resulting in a batch risk label that includes an eigenvector decomposition description.

7. A method for quality traceability throughout the entire production process of oral liquid drugs according to claim 6, comprising generating differentiated traceability record packages based on batch risk marking, including: Generate a first-level process-level operation compliance record that includes the comparison results of process parameter specification limits; Generate a second-layer process deviation analysis record that includes process deviation feature vectors, predicted stability acceleration decay slope, and threshold comparison results; Based on the operation records and operator information associated with the main sources of risk, a third layer of risk tracing labeling is generated, ultimately resulting in a differentiated traceability record package.

8. The method for quality traceability throughout the entire production process of oral liquid drugs according to claim 1, wherein release decision support suggestions are added to the traceability record package, including: Generate supplementary testing recommendations for assessing actual antioxidant consumption; Generate recommendations to include the corresponding batches in the enhanced stability retention program; The relevant testing recommendations and suggestions for strengthening the stability retention plan are pushed through the data interface, ultimately obtaining release decision support recommendations.

9. The method for quality traceability of the entire production process of oral liquid drugs according to claim 1, wherein the process deviation feature vectors marked as corresponding to regular batches are included in the historical sample database to update the judgment threshold, including: The process deviation feature vectors of regular batches and the stability retention data continuously collected after release are automatically stored in the historical sample library. The distribution of qualified individuals was re-statistically analyzed based on the expanded historical sample database; The quantiles are recalculated based on the updated distribution of the qualified population, and the updated judgment threshold is finally obtained.

10. A quality traceability system for the entire production process of oral liquid drugs, characterized in that, include: The system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement a method for quality traceability throughout the entire production process of oral liquid pharmaceuticals according to any one of claims 1-9.

Images