A medicine dosage form optimization direction recommendation method and system based on market feedback data

Abstract

The application belongs to the technical field of pharmaceutical preparation data processing, and discloses a medicine dosage form optimization direction recommendation method and system based on market feedback data. The method is related to market feedback of different dosage forms, same user dosage form conversion records and corresponding batch detection data, constructs user internal matching samples, extracts feedback events and pharmacological negative control events; confirms dosage form problems and confidence levels through user internal problem differences, negative control correction and batch performance direction verification; maps the confirmation results to key quality attributes, generates candidate prescriptions and process parameter combinations, corrects constraint relationships through performance prediction and trial production detection, and outputs target dosage forms, parameter ranges and verification items. The application can reduce false attribution caused by non-dosage form factors, and improve the implementability and verification efficiency of the dosage form optimization direction.

Description

Technical Field

[0001] This invention relates to the fields of pharmaceutical formulation and data processing technology, specifically to a method and system for recommending drug dosage form optimization directions based on market feedback data. Background Technology

[0002] Drug dosage form is a crucial factor influencing drug administration convenience, disintegration properties, dissolution properties, release behavior, storage stability, and patient medication adherence. After a drug is marketed, patient reviews, complaint records, reasons for returns / exchanges, medication adherence records, adverse reaction reports, and follow-up records typically include information such as difficulty swallowing, delayed onset of action, unusual odor, sedimentation, clumping, damaged coating, or inconvenience of use. This market feedback reflects the suitability of drug dosage forms for different user groups from the actual usage perspective, providing a data foundation for continuous improvement of drug formulations, manufacturing processes, and dosage form structures.

[0003] Existing pharmaceutical quality management technologies have begun to utilize big data for comprehensive analysis of R&D, production, supply chain, and market evaluation data. For example, Chinese patent application CN116597939A discloses a big data-based pharmaceutical quality control management analysis system and method. This system collects pharmaceutical R&D data, pharmaceutical production sensor data, supply chain data, and evaluation feedback data. It performs noise reduction, feature extraction, and deep neural network analysis on the data and displays quality control strategies and management solutions. While this solution can improve the processing efficiency of multi-source pharmaceutical quality data, it primarily focuses on pharmaceutical quality control management. It does not establish a pairing relationship between different dosage forms of the same active ingredient before and after user conversion, nor does it provide a method for further attributing market feedback issues to specific dosage form attributes and converting them into prescription parameters or preparation process parameters.

[0004] Market feedback data is characterized by its dispersed sources, inconsistent presentation formats, and complex interplay of influencing factors. The same feedback issue may stem from dosage form performance factors such as dosage size, disintegration rate, dissolution behavior, particle size, or viscosity, or it may be caused by inherent pharmacological effects of the active ingredient, patient baseline condition, changes in co-administered medications, or fluctuations in individual production batches. If dosage form optimization is determined solely based on the number of complaints, evaluation frequency, or sentiment bias, it is easy to misidentify inherent reactions of the active ingredient or changes in user condition as dosage form problems, leading to ineffective adjustments to parameters such as excipients, compression pressure, coating weight gain, and homogenization strength.

[0005] When real-world data is used for drug development and evaluation, appropriate controls, bias control, and suitable statistical analysis are necessary to improve the reliability of the evidence. However, existing technologies typically implement market feedback analysis, production batch quality testing, and formulation optimization separately, lacking a mechanism for continuous verification of feedback before and after dosage form conversion for the same user, pharmacological negative control events, and corresponding batch physicochemical test results. Furthermore, existing formulation optimization methods often rely directly on laboratory formulation data and process test data, making it difficult to determine whether problems in market feedback are indeed caused by dosage form attributes.

[0006] Therefore, there is a need for a technical solution that can link market feedback, dosage form conversion records, and batch testing data, thereby eliminating user differences, the combined influence of active ingredients, and occasional batch fluctuations, and mapping confirmed dosage form issues to key quality attributes, formulation parameters, and preparation process parameters, in order to improve the reliability and feasibility of drug dosage form optimization. Summary of the Invention

[0007] The purpose of this invention is to provide a method and system for recommending drug dosage form optimization based on market feedback data, so as to identify dosage form problems from market feedback data affected by user differences, inherent reactions of active ingredients and production batch fluctuations, and generate a range of prescription parameters and preparation process parameters that can be tested and detected.

[0008] To achieve the above-mentioned technical objectives, the present invention provides the following technical solution.

[0009] In a first aspect, the present invention provides a method for recommending drug dosage form optimization directions based on market feedback data, comprising the following steps:

[0010] S1. Obtain one or more of the following: market feedback data for each dosage form of the target drug, dosage form conversion records for the same user, and disintegration, dissolution, particle size, viscosity, and stability test data for the corresponding production batch and the dosage form, and associate them according to user, active ingredient, dosage specification, indication, production batch, and medication stage;

[0011] S2. Form a paired sample within the user based on market feedback data before and after the same user's dosage form conversion and extract feedback events; select negative control events from systemic reactions that are present in all dosage forms and caused by the pharmacological effects of the active ingredients; determine batch performance deviations based on the deviations of the test data from the registration standard control range or historical stable intervals;

[0012] S3. For each feedback issue, determine the intra-user issue difference, subtract the difference of the negative control event in the same period, and obtain the corrected issue difference; when the batch performance deviation and the corrected issue difference meet the preset direction correspondence, confirm the dosage form attribution and obtain the dosage form issue and confidence level;

[0013] S4. Map the confirmed dosage form issues to key quality attributes. Based on the constraints between key quality attributes and formulation parameters and preparation process parameters, generate candidate parameter combinations that meet the constraints of active ingredient stability, route of administration, bioavailability and drug quality standards, and predict the disintegration, dissolution, release, particle size, viscosity, dispersion or stability performance corresponding to each candidate parameter combination.

[0014] S5. Select candidate parameter combinations that meet the preset requirements for predicted performance improvement and confidence level, and conduct trial production. Correct the constraint relationship based on the deviation between the measured performance and the predicted performance, and output the target dosage form, prescription parameter range, preparation process parameter range, and verification items.

[0015] Specifically, in step S1: the market feedback data includes one or more of the following: drug evaluation texts, complaint records, records of reasons for drug return or exchange, medication adherence records, adverse reaction reports, and follow-up records;

[0016] The dosage form conversion record includes at least the user identifier, dosage form before conversion, dosage form after conversion, conversion time, dosage, and production batch identifier;

[0017] The market feedback data is associated with the corresponding production batch test data based on the production batch identifier, and the association is verified based on the generation time, dosage form conversion time, and medication stage of the market feedback data.

[0018] Specifically, in step S2, forming the user-to-user pairing sample includes: screening market feedback data generated by the same user using the first and second dosage forms of the target drug within a preset pairing time range;

[0019] Verify that the active ingredients, dosage specifications, indications, and dosing regimens corresponding to the first and second dosage forms are consistent;

[0020] Exclude dosage form conversion records that involve changes in active ingredients, dosage specifications, indications, or combination therapy regimens during the dosage form conversion period;

[0021] The market feedback data before and after the dosage form conversion, which meet the verification criteria, will be used to form a user-matched sample.

[0022] Specifically, in step S2, extracting the feedback event includes: standardizing the synonyms, negations, degree expressions, and time expressions in the market feedback data;

[0023] Extract feedback events from standardized market feedback data, which consist of the problem type, stage of occurrence, severity, source of evidence, and dosage form attributes involved.

[0024] Duplicate feedback events are merged based on user identifier, medication stage, time of occurrence of feedback event, and content of feedback event;

[0025] The dosage form attributes include one or more of the following: dosage form size, surface coating, disintegration properties, dissolution properties, release properties, particle state, viscosity state, dispersion state, and storage stability.

[0026] Specifically, in step S2, the negative control event is selected according to the following conditions:

[0027] The negative control event is a systemic reaction related to the active ingredient recorded in the drug instructions or a pre-set pharmacological knowledge base;

[0028] The negative control events are recorded in each of the comparison dosage forms of the target drug;

[0029] The difference in the occurrence of the negative control event among the various comparison dosage forms is lower than a preset dosage form difference threshold.

[0030] The correlation between the negative control event and the data on disintegration, dissolution, particle size, viscosity, and stability is lower than a preset correlation threshold.

[0031] Specifically, in step S2, determining the batch performance deviation includes comparing the disintegration, dissolution, particle size, viscosity and stability test values ​​of each production batch with the corresponding registration standard control range;

[0032] When the registration standard does not specify the corresponding control range, the test value is compared with the historical stable range determined based on the historical qualified batches of the target drug;

[0033] Based on the deviation index, direction of deviation, and degree of deviation of the detected value relative to the registered standard control range or historical stable interval, a batch performance deviation record for the corresponding production batch is generated.

[0034] Specifically, in step S3: the differences in user-related issues are determined based on the occurrence status, frequency, and severity of feedback events before and after the dosage form change;

[0035] The differences in negative control events during the same treatment period are determined based on the occurrence status, frequency, and severity of the negative control events.

[0036] The corrected problem difference is obtained by subtracting the impact of the difference in the negative control event during the same period from the difference in the problem within the user;

[0037] When multiple negative control events exist, the correction priority of each negative control event is determined based on the sample size, record completeness, and reliability of the evidence source.

[0038] Specifically, in step S3: the preset direction correspondence is a pre-established correspondence between feedback issues, key quality attributes, and the risk change direction of key quality attribute detection values;

[0039] When the difference in the correction problem indicates that the corresponding dosage form problem has worsened, and the detection value of the corresponding key quality attribute changes in the risk direction of deviating from the target control range, it is determined that the batch performance deviation and the difference in the correction problem satisfy a preset directional correspondence.

[0040] When the difference in the correction problem indicates that the corresponding dosage form problem has been alleviated, and the detection value of the corresponding key quality attribute returns to the target control range, it is determined that the batch performance deviation and the difference in the correction problem satisfy a preset directional correspondence.

[0041] If any of the above conditions are not met, it is not confirmed that the corresponding feedback event was caused by the dosage form attribute.

[0042] Specifically, in step S3: the confidence level of the dosage form problem is determined based on the number of user-paired samples that form the difference in the correction problem, the degree of difference in the correction problem, the number and proportion of production batches that meet the preset directional correspondence, and the reliability of the evidence source of the feedback event;

[0043] According to the preset confidence classification conditions, the confidence level is divided into high confidence level, medium confidence level or low confidence level.

[0044] Specifically, in step S4: step S4 only generates candidate parameter combinations for dosage form problems that reach a medium or high confidence level.

[0045] The key quality attributes include one or more of the following: disintegration time, dissolution profile, release rate, particle size distribution, viscosity, dispersion uniformity, and storage stability.

[0046] The formulation parameters include one or more of the following: type and dosage of disintegrant, type and dosage of binder, coating material and coating weight gain, type and dosage of suspending agent and type and dosage of release modifier.

[0047] The preparation process parameters include one or more of the following: granulation endpoint, tableting pressure, coating temperature, coating speed, particle size, homogenization strength, and drying conditions.

[0048] The constraint relationship is used to characterize the relationship between the adjustment direction and range of the prescription parameters or preparation process parameters and the change direction of the corresponding key quality attributes.

[0049] Specifically, in step S5: the predicted performance improvement amount is the degree to which the predicted performance corresponding to the candidate parameter combination approaches the preset target range relative to the baseline performance, and the baseline performance is the historical stability performance or the measured performance of the most recent qualified production batch corresponding to the current prescription parameters and preparation process parameters.

[0050] When the candidate parameter combination corresponds to multiple key quality attributes, the improvement in the predictive performance of each key quality attribute is compared with the corresponding preset improvement threshold.

[0051] The prototype samples were prepared according to the candidate parameter combinations, and the prototype samples were subjected to performance tests corresponding to the key quality attributes.

[0052] When the measured performance reaches the preset target range, the recommendation priority of the corresponding candidate parameter combination is increased, and the corresponding constraint relationship is retained;

[0053] When the measured performance does not reach the preset target range, or the deviation between the measured performance and the predicted performance exceeds the preset allowable deviation, the candidate range of the corresponding formulation parameters or preparation process parameters is adjusted, and the confidence level of the corresponding constraint relationship is reduced.

[0054] Based on the adjusted constraint relationships, candidate parameter combinations are regenerated until a candidate parameter combination whose measured performance meets the preset target range is obtained, or the preset number of trial productions is reached.

[0055] Secondly, the present invention further provides a drug dosage form optimization direction recommendation system based on market feedback data, used to implement the drug dosage form optimization direction recommendation method based on market feedback data described in the first aspect, the system comprising:

[0056] The data association module is used to associate market feedback data for various dosage forms of the target drug, dosage form conversion records of the same user, and test data of the corresponding production batches.

[0057] The dosage form attribution module is used to generate in-user paired samples, extract feedback events, perform negative control event correction and batch performance deviation verification, and determine dosage form issues and their confidence levels.

[0058] The parameter recommendation module is used to generate candidate parameter combinations based on dosage form issues confirmed by dosage form attribution, predict the formulation performance corresponding to the candidate parameter combinations, and output the dosage form optimization direction.

[0059] The trial production correction module is used to correct the constraint relationship between key quality attributes and formulation parameters and preparation process parameters based on the measured performance of the trial production sample.

[0060] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein when the processor executes the computer program, it implements the drug dosage form optimization direction recommendation method based on market feedback data as described in the first aspect.

[0061] This invention forms paired samples within the same user group by analyzing market feedback before and after dosage form conversion, reducing interference from different user baseline states and feedback habits. It uses pharmacological reactions that co-occur in different dosage forms and have low correlation with dosage form performance as negative control events to correct for differences in user-specific issues. Furthermore, it performs directional verification by comparing the corrected issues with corresponding production batches' disintegration, dissolution, particle size, viscosity, or stability deviations. Only issues meeting preset directional correspondences are confirmed as having dosage form attribution, thus distinguishing between inherent reactions of active ingredients, changes in user status, occasional batch anomalies, and dosage form performance problems. Based on this, the invention maps confirmed dosage form issues to key quality attributes, generates candidate parameter combinations based on the constraints between key quality attributes and formulation parameters and preparation process parameters, and corrects these constraints using the measured performance of trial samples. This creates a closed loop between market feedback analysis, dosage form issue confirmation, parameter recommendation, and trial production verification, which helps reduce ineffective formulations or process adjustments and improves the accuracy, feasibility, and verification efficiency of dosage form optimization schemes. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the drug dosage form optimization direction recommendation method based on market feedback data of the present invention;

[0063] Figure 2 This is a schematic diagram illustrating the correlation between market feedback data, dosage form conversion records, and production batch testing data of this invention.

[0064] Figure 3 This is a schematic diagram of the three-level attribution process for dosage form issues in this invention;

[0065] Figure 4 This is a schematic diagram illustrating the process of generating candidate parameter combinations, prototyping and testing, and correcting constraint relationships in this invention. Detailed Implementation

[0066] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Equivalent substitutions or conventional modifications made by those skilled in the art to the following embodiments without departing from the technical concept of the present invention are all applicable to the present invention.

[0067] like Figure 1As shown, the present invention provides a method for recommending drug dosage form optimization directions based on market feedback data, including the association of market feedback data, dosage form conversion records and production batch test data, the formation of user-matched samples and feedback events, the attribution confirmation of dosage form problems, the generation of candidate parameter combinations, and the correction of trial production testing and constraint relationships.

[0068] like Figure 2 As shown, for the same user, market feedback data, medication stage, and corresponding production batch testing data before and after dosage form conversion are associated to form a user-paired sample that can reflect the changes in problems of the same user under different dosage form usage conditions.

[0069] like Figure 3 As shown, the confirmation of dosage form issues includes, in sequence, determining the differences in issues within the user group, correcting for differences in contemporaneous negative control events, and verifying the direction of batch performance deviations. Only when the corrected differences in issues and the batch performance deviations meet the preset directional correspondence is the attribution relationship between the corresponding feedback event and the dosage form attribute confirmed.

[0070] like Figure 4 As shown, dosage form issues identified through dosage form attribution are mapped to key quality attributes, and candidate parameter combinations are generated based on the constraint relationships between key quality attributes and formulation parameters and preparation process parameters. Subsequently, the performance of the candidate parameter combinations is predicted, tested, and verified, and the constraint relationships are corrected based on the deviation between the measured performance and the predicted performance.

[0071] I. Terminology Explanation

[0072] In this embodiment, "dosage forms" of the target drug refer to drug dosage forms with the same active ingredient, indications, and routes of administration, but different physical forms, drug release structures, dispersion states, or methods of use. Examples include ordinary tablets and dispersible tablets, ordinary capsules and enteric-coated capsules, ordinary release tablets and sustained-release tablets, and oral suspensions and oral solutions.

[0073] When the dosage forms being compared have different dosage strengths, the total daily dose should be converted according to the dosing regimen, and paired samples within the user group should only be formed if the total daily dose is equivalent. For dosage forms where different release methods lead to changes in dosing frequency, the dosing frequency and total dose per unit time before and after the conversion should also be verified.

[0074] Market feedback data refers to data generated by patients, medical personnel, pharmaceutical companies, after-sales service providers, or pharmacovigilance systems after a target drug enters the market and reflects its usage. This market feedback data includes one or more of the following: drug evaluation texts, complaint records, records of reasons for drug returns or exchanges, medication adherence records, adverse reaction reports, and follow-up records.

[0075] Feedback events refer to structured problem records extracted from one or more market feedback data points. A feedback event must include at least the problem type, stage of occurrence, severity, source of evidence, and relevant dosage form attributes. For example, "tablets are difficult to swallow," "prolonged onset time after administration," "suspension settles after standing," "difficult to redisperse after shaking," and "damaged tablet coating" can all form corresponding feedback events.

[0076] A negative control event is a systemic reaction caused by the pharmacological action of the active ingredient, recorded in all comparable dosage forms of the target drug, and with a low correlation to data on disintegration, dissolution, particle size, viscosity, and stability. Negative control events are used to characterize common differences experienced by the same user before and after dosage form switching, caused by the inherent effects of the active ingredient, changes in underlying disease, or changes in overall feedback patterns.

[0077] Batch performance deviation refers to the deviation of a particular production batch of a target drug from the control range of the registration standard or the historical stable range in terms of disintegration, dissolution, particle size, viscosity, or stability test values, the direction of deviation, and the degree of deviation.

[0078] Pre-defined directional correspondences refer to the pre-established correspondence between the direction of change in feedback issues and the direction of risk change in the corresponding key quality attribute detection values. For example, an aggravated problem of prolonged onset time corresponds to a decrease in dissolution at a specified time point; an aggravated problem of tablet disintegration corresponds to a prolonged disintegration time; and an aggravated problem of suspension sedimentation corresponds to an increase in particle size, a decrease in sedimentation volume ratio, or a deviation of viscosity from the appropriate range.

[0079] Prescription parameters refer to parameters that limit the composition, proportions, and usage of functional excipients of the target dosage form, including one or more of the following: type and amount of disintegrant, type and amount of binder, coating material and coating weight gain, type and amount of suspending agent, and type and amount of release modifier.

[0080] Preparation process parameters refer to parameters that can be adjusted during the preparation of the target dosage form and affect key quality attributes, including one or more of the following: granulation endpoint, tableting pressure, coating temperature, coating speed, particle size, homogenization strength, and drying conditions.

[0081] II. Data Processing Platform

[0082] This invention can be deployed in pharmacovigilance platforms, formulation development servers, quality management servers, or other electronic devices with data processing capabilities in pharmaceutical manufacturing enterprises.

[0083] The platform includes at least:

[0084] 1. Market feedback database, used to store drug evaluation texts, complaint records, records of reasons for drug returns and exchanges, adverse reaction reports, and follow-up records;

[0085] 2. Dosage form conversion record database, used to store user ID, dosage form before conversion, dosage form after conversion, conversion time, dosage specification, dosing regimen and production batch ID;

[0086] 3. Batch testing database, used to store disintegration, dissolution, particle size, viscosity and stability test data for each production batch;

[0087] 4. Formulation knowledge base, used to store feedback issues, key quality attributes, formulation parameters, preparation process parameters and their directional correspondences;

[0088] 5. Trial production and verification database, used to store candidate parameter combinations, predicted performance, measured performance of trial production samples, and records of constraint relationship corrections.

[0089] To protect user information, market feedback data is de-identified before being included in this method. User names, contact information, ID card numbers, and other direct identifiers are not included in this method; only anonymous user identifiers that can be used to correlate data before and after dosage form conversion are retained.

[0090] III. Overall Method Flow

[0091] like Figure 1 As shown, the method of the present invention includes steps S1 to S5.

[0092] (I) Step S1: Link market feedback data, dosage form conversion records, and production batch testing data.

[0093] Obtain one or more of the following: market feedback data for each dosage form of the target drug, dosage form conversion records for the same user, and disintegration, dissolution, particle size, viscosity, and stability test data for the corresponding production batch and the dosage form. Then, associate these data with the user, active ingredient, dosage specification, indication, production batch, and stage of administration.

[0094] 1. Standardized processing of market feedback data

[0095] Market feedback data can be structured records or natural language text.

[0096] For structured records, fields such as drug code, dosage form code, user identifier, feedback time, problem type, and severity can be read directly.

[0097] For natural language text, character normalization, synonym merging, negation expression recognition, degree expression recognition, and time expression recognition are performed in sequence.

[0098] For example, expressions such as "the pill is too big to swallow", "difficulty swallowing", and "the pill is stuck in the throat" are uniformly classified into the "difficulty swallowing" problem type; expressions such as "no nausea" and "no dizziness" are identified as negative events; and expressions such as "slight sedimentation", "significant sedimentation", and "complete clumping" are mapped to different degrees of severity.

[0099] This implementation method can use a pre-established rule dictionary and text processing rules to extract feedback events. The rule dictionary includes a question type dictionary, a dosage form attribute dictionary, a degree dictionary, a negation dictionary, and a time stage dictionary.

[0100] In other embodiments, a trained text classification model can be used to initially classify market feedback texts, but the classification results should be uniformly output as structured feedback events including problem type, stage of occurrence, severity, source of evidence, and dosage form attributes. This invention does not require the use of a specific text classification model.

[0101] Standardized market feedback records should include at least:

[0102] User anonymity identifier, drug code, active ingredient code, dosage form code, dosage specification, indication, dosing regimen, feedback time, production batch identifier, problem type, stage of occurrence, severity, source of evidence, and dosage form attributes involved.

[0103] 2. Verification of dosage form change records

[0104] Dosage form conversion records should include at least the user ID, dosage form before conversion, dosage form after conversion, conversion time, dosage, and production batch ID.

[0105] Before generating in-user paired samples, the following verifications are performed on the conversion records:

[0106] First, the active ingredients of the drugs are the same before and after the conversion;

[0107] Second, the indications for the drugs before and after the conversion are the same;

[0108] Third, the dosage and specifications of the drugs are the same before and after the conversion, or the total daily dosage is the same after conversion according to the dosing regimen;

[0109] Fourth, the route of administration for the drugs remains the same before and after the conversion;

[0110] Fifth, the main combination drug was not changed simultaneously during the dosage form conversion;

[0111] Sixth, no changes in disease status occurred during the dosage form switch that would significantly affect the feedback results.

[0112] When a user experiences hospitalization, surgery, acute infection, significant dose adjustment, or new combination therapy that affects gastrointestinal status or drug absorption during dosage form conversion, the corresponding conversion record will be marked as an interference record and will not be used to form a paired sample within the user.

[0113] 3. Acquisition of production batch testing data

[0114] For tablets or capsules, the test data may include disintegration time, dissolution amount at different time points, dissolution curve, hardness, friability, content, and stability data.

[0115] For granules or dry suspensions, the test data may include particle size distribution, reconstitution time, sedimentation volume ratio, redispersibility, viscosity, and stability data.

[0116] For oral liquid formulations, the test data may include particle size, viscosity, sedimentation state, redispersibility, content uniformity, and stability data.

[0117] Although the data platform can store disintegration, dissolution, particle size, viscosity, and stability test data, when processing specific feedback issues, it only calls upon the test indicators relevant to the dosage form and the corresponding problem. For example, for the issue of prolonged tablet onset time, it calls upon disintegration and dissolution test data; for the issue of suspension sedimentation, it calls upon particle size, viscosity, sedimentation volume ratio, and redispersibility data.

[0118] 4. Establishing related records

[0119] like Figure 2 As shown, the market feedback data before and after the dosage form conversion for the same user is associated with the user's anonymous identifier as the main index; the feedback event is associated with the corresponding production batch test data with the production batch identifier as the batch index.

[0120] When the market feedback data directly contains the production batch identifier, a direct correspondence is established between the feedback event and the batch test data.

[0121] When market feedback data does not directly record the production batch identifier, but records the drug traceability code, prescription dispensing time, packaging scanning record, or inventory release record, the production batch can be determined by parsing the drug traceability code, or by determining the corresponding production batch based on the dispensing time and inventory release record.

[0122] Feedback records that cannot be reliably identified for a production batch can be used to identify differences in issues within a user group, but they are not directly used to verify the direction of batch performance deviations, or they may reduce the reliability of the evidence source.

[0123] After the association is completed, a linked record can be represented as:

[0124]

[0125] in, Indicates associated records; Indicates user anonymity; Indicate dosage form; Indicates dosage specifications and dosing regimen; Indicates the production batch; Indicates the stage of medication and the corresponding time; Indicates a feedback event; This represents the set of test data for the corresponding production batch.

[0126] (ii) Step S2: Forming paired samples within the user and identifying negative control events and batch performance bias.

[0127] Paired samples within the same user group were formed using market feedback data before and after the dosage form conversion, and feedback events were extracted. Negative control events were selected from systemic reactions that occurred in all dosage forms and were caused by the pharmacological effects of the active ingredients. Batch performance deviations were determined based on the deviations of the test data from the control range of the registration standard or the historical stable range.

[0128] 1. Formation of user-matched samples

[0129] For each user who switches dosage forms, separate observation windows are set up for the period before and after the switch.

[0130] The observation window before the conversion can be set to 7 to 30 days before the conversion date; the observation window after the conversion can be set to 3 to 30 days after the conversion is completed and the adaptation period has ended.

[0131] The adaptation period is used to exclude short-term unstable feedback caused by changes in the drug's appearance, packaging, or method of administration. The length of the observation window can be adjusted based on the drug's dosing cycle, the pharmacokinetic characteristics of the active ingredient, and the density of market feedback data.

[0132] User-matched samples are formed in the following way:

[0133] Filter market feedback data generated by the same user using the first and second dosage forms of the target drug within a preset pairing time range;

[0134] Verify that the active ingredients, dosage specifications, indications, and dosing regimens of the first and second dosage forms are consistent.

[0135] Exclude dosage form conversion records where active ingredient changes, dosage specification adjustments, indication changes, or combination therapy regimen changes occurred during the dosage form conversion period;

[0136] The market feedback data before and after the dosage form conversion, which meet the verification criteria, will be used to form a user-matched sample.

[0137] For a certain issue type, if it did not exist before the conversion but exists after the conversion, it is recorded as a newly occurring issue; if it existed before the conversion but does not exist after the conversion, it is recorded as a disappeared issue; if it existed before and after the conversion, the frequency and severity of occurrence are compared; if it does not exist before and after the conversion, the corresponding issue is not included in the issue difference calculation for that user.

[0138] Duplicate feedback events generated by the same user during the same medication period can be merged based on user identification, time of occurrence, and feedback content. For example, if the same user submits two consecutive records within 24 hours, namely "pill too big" and "difficulty swallowing," they can be merged into a single "difficulty swallowing" feedback event, retaining the higher severity level.

[0139] 2. Structured representation of feedback events

[0140] Feedback events can be represented as:

[0141]

[0142] in, Indicates a feedback event; Indicate the question type; Indicates the stage of occurrence; Indicates the degree of severity; Indicate the source of the evidence; This indicates that dosage form attributes are involved.

[0143] The severity can be divided into levels 0 to 3:

[0144] Level 0 indicates that it did not occur;

[0145] Level 1 indicates a minor issue that does not affect continued use;

[0146] Level 2 indicates that it affects normal medication administration or requires adjustment of the method of use;

[0147] Level 3 indicates that medication should be discontinued, returned, or changed to a different dosage form, or that medical treatment is required.

[0148] Evidence sources can be categorized into levels of reliability. For example, a single patient self-report corresponds to Level 1 evidence; repeated patient feedback or pharmacist follow-up verification corresponds to Level 2 evidence; and medical staff records, physical examinations of returned or exchanged medications, or laboratory tests correspond to Level 3 evidence.

[0149] 3. Selection of negative control events

[0150] Negative control events are selected based on the following criteria:

[0151] First, the event is a systemic reaction related to the active ingredient as recorded in the drug instructions or a pre-set pharmacological knowledge base;

[0152] Second, the events are recorded in each of the comparison dosage forms of the target drug;

[0153] Third, the difference in the occurrence of the event among the various dosage forms to be compared is lower than a preset dosage form difference threshold;

[0154] Fourth, the correlation between the event and the detection data of disintegration, dissolution, particle size, viscosity and stability is lower than the preset correlation threshold.

[0155] For example, if an active ingredient is known to cause mild dizziness or fatigue, and such reactions have been recorded in both regular tablets and dispersible tablets, and the difference in their occurrence is below a preset threshold and is not significantly related to tablet disintegration time and dissolution, then mild dizziness or fatigue can be used as a negative control event.

[0156] In one implementation, Spearman's rank correlation coefficient is used to evaluate the correlation between the proportion of negative control events and various detection data. When the absolute value of each correlation coefficient is below 0.2, the correlation is determined to be below a preset correlation threshold. Depending on the data scale and business needs, the correlation threshold can also be set to 0.15 to 0.30.

[0157] When there are multiple negative control events that meet the criteria, the correction priority or correction weight is determined based on the sample size, record completeness, and reliability of the evidence source for each negative control event.

[0158] 4. Determination of batch performance deviation

[0159] The test values ​​for each production batch are compared with the corresponding control range of the registration standard. When the registration standard does not specify a corresponding control range, the test values ​​are compared with the historical stable interval determined based on the historical qualified batches of the target drug.

[0160] When the detection indicator has a target range At that time, production batch The The batch performance deviation of the test index can be determined according to the following formula:

[0161]

[0162] in, Indicates production batch The Standardized deviation of the test indicators; Indicates the corresponding detection value; Indicates the lower limit of the target interval; This indicates the upper limit of the target range.

[0163] when When the value is positive, it indicates that the detected value is higher than the target range; when... When the value is negative, it indicates that the detected value is lower than the target range; its absolute value indicates the degree of deviation.

[0164] When the registration standard only specifies a one-sided limit, the limit and the median of historical qualified batches can be used as a reference standard.

[0165] When the registration standard does not specify the corresponding control range, test data from no less than 20 historical qualified production batches can be read, and the historical stable interval can be constructed using the 10th percentile and the 90th percentile. For data that is approximately normally distributed, the historical stable interval can also be constructed by adding or subtracting two standard deviations from the historical mean.

[0166] The batch performance deviation record should include at least the test index, test value, reference range, deviation direction, deviation degree, and test time.

[0167] (III) Step S3: Identify the discrepancies in the correction problem and confirm the dosage form attribution.

[0168] For each feedback issue, the user-internal issue difference is determined, and the difference of the negative control event in the same period is deducted to obtain the correction issue difference; when the batch performance deviation and the correction issue difference meet the preset direction correspondence, the dosage form attribution is confirmed, and the dosage form issue and confidence level are obtained.

[0169] like Figure 3 As shown, this step sequentially includes determining the differences in user-related issues, correcting for negative control events, and verifying the direction of batch performance deviations.

[0170] 1. Determining the differences in user-related issues

[0171] For users and question types The differences in user-related issues are determined based on the occurrence status, frequency, and severity of feedback events before and after the dosage form change.

[0172] In one implementation, user-specific issues differ. satisfy:

[0173]

[0174] in, Indicates user According to the type of problem Differences in user-related issues; and These indicate the severity of the problems before and after the dosage form change; and These represent the frequency of the problem after standardization based on the observation duration; Indicates the severity weight; This indicates the frequency weighting.

[0175] In this embodiment, it is possible to set , For problems such as difficulty swallowing and tablet breakage, which are mainly characterized by a single severe feedback, the severity weight can be increased; for recurring usage problems such as sedimentation and clumping, the frequency weight can be increased.

[0176] when When, it indicates that the corresponding problem is aggravated after conversion; when When the value is less than the preset minimum difference threshold, it indicates that the corresponding problem has been alleviated; when the absolute value is less than the preset minimum difference threshold, it is determined that the user has not experienced a valid change in problem.

[0177] The differences in intra-user issues among multiple users with the same dosage form conversion direction are summarized to obtain the mean or median of the intra-user issues for the corresponding feedback issues.

[0178] 2. Determination of differences in concurrent negative control events

[0179] Based on the observation windows before and after the conversion used for the target feedback problem, the differences in negative control events for the same user are determined.

[0180] Let the first The intra-user difference of each negative control event is: When multiple negative contrast events exist, the user Comprehensive negative control event differences satisfy:

[0181]

[0182] in, Indicates the number of negative control events; Indicates the first The corrected weights for each negative control event are given, and the sum of the corrected weights is 1.

[0183] The correction weights can be determined based on the sample size, record completeness, and reliability of the evidence source for the negative control event. Negative control events with larger sample sizes, more complete records, and reliable evidence sources have higher correction weights.

[0184] 3. Determining the discrepancies in the correction problem

[0185] The corrected issue difference is obtained by subtracting the impact of the difference in negative control events during the same period from the issue difference within the user base:

[0186]

[0187] in, Indicates user According to the type of problem Differences in correction issues; Indicates differences in user-defined issues; Indicates the difference in overall negative control events during the same period; This represents the negative control correction coefficient.

[0188] When the differences in user-related issues and the differences in negative control events are measured using the same severity and frequency scales, the negative control correction coefficient can be set to 1.

[0189] When the data densities or dimensions of the two are different, the negative control correction coefficient can be determined using user records that have not undergone dosage form conversion, so that the mean difference of the problem after negative control correction for users who have not undergone dosage form conversion is close to 0.

[0190] 4. Preset direction correspondence

[0191] The preset directional correspondence is stored in the formulation knowledge base and is used to characterize the correspondence between feedback issues, key quality attributes, and the direction of risk changes in key quality attribute detection values.

[0192] Table 1 lists some examples to illustrate how the direction correspondence is set.

[0193] Table 1. Examples of the Correspondence Between Feedback Issues and Key Quality Attribute Directions

[0194]

[0195] When the correction problem difference indicates that the corresponding dosage form problem has worsened, and the detection value of the corresponding key quality attribute changes in the risk direction of deviating from the target control range, it is determined that the batch performance deviation and the correction problem difference meet the preset directional correspondence.

[0196] When the correction problem difference indicates that the corresponding dosage form problem has been alleviated, and the detection value of the corresponding key quality attribute returns to the target control range, it is also determined that the batch performance deviation and the correction problem difference meet the preset directional correspondence.

[0197] If any of the above conditions are not met, it is not confirmed that the corresponding feedback event was caused by the dosage form attribute.

[0198] For example, if the correction of the problem shows that the onset time has been prolonged, but the dissolution amount at the specified time point of the corresponding production batch has increased and is within the target range, then the preset direction correspondence is not met, and the feedback problem is not identified as a dosage form problem caused by dissolution performance.

[0199] When the same feedback issue is associated with multiple production batches, the number of production batches that meet the preset directional correspondence and their proportion are counted. When the proportion reaches a preset percentage, the dosage form attribution is confirmed. The preset percentage can be between 60% and 80%.

[0200] 5. Determination of confidence level

[0201] The confidence level for dosage form issues is determined based on the following information:

[0202] The number of paired samples within the user group that generated the difference in the correction problem;

[0203] Correcting the degree of difference in the problem;

[0204] The number and percentage of production batches that meet the preset directional correspondence;

[0205] The reliability of the evidence sources for the feedback event.

[0206] In one implementation, confidence scoring for dosage form issues satisfy:

[0207]

[0208] in, Indicate the problem type The corresponding confidence score; The score represents the number of paired samples within a user. Indicates the difference score for the correction problem; This indicates the scoring of production batches that are consistent in direction. Indicate the source of evidence score; , , and These represent the corresponding weights, and .

[0209] In this embodiment, it can be set , , , .

[0210] when When the confidence level is not lower than 0.75, the corresponding dosage form issue is classified as high confidence; when... A value not lower than 0.55 and lower than 0.75 is classified as a medium confidence level; when... A value below 0.55 is classified as a low confidence level.

[0211] Candidate parameter combinations are generated only for dosage form issues that reach a medium or high confidence level. Dosage form issues with a low confidence level remain pending new market feedback data or production batch testing data, but will not directly trigger adjustments to formulation parameters and preparation process parameters.

[0212] The proportion of production batches with consistent direction refers to the ratio of the number of production batches that have passed the direction verification to the number of production batches whose critical quality attributes deviate from the risk direction and whose correction problem differences have reached the effective change conditions. Production batches whose critical quality attributes are within the target control range and whose correction problems have not changed effectively are considered as stable control batches and are not included in the calculation of the proportion.

[0213] (iv) Step S4: Generate candidate parameter combinations and predict formulation performance

[0214] The identified dosage form issues are mapped to key quality attributes. Based on the constraints between key quality attributes and formulation parameters and preparation process parameters, candidate parameter combinations that meet the constraints of active ingredient stability, route of administration, bioavailability and drug quality standards are generated. The disintegration, dissolution, release, particle size, viscosity, dispersion or stability properties corresponding to each candidate parameter combination are predicted.

[0215] 1. Mapping of Dosage Form Issues to Key Quality Attributes

[0216] The mapping relationship between dosage form issues and key quality attributes can be pre-established based on drug registration data, formulation research data, process validation data, historical deviation investigation records, and formulation knowledge base.

[0217] For example:

[0218] The extended onset time is mapped to the disintegration time limit, the amount of dissolution at a specified time point, and the dissolution profile;

[0219] Difficulty swallowing is reflected in tablet diameter, tablet thickness, surface condition, and disintegration properties;

[0220] Suspension sedimentation is mapped to particle size distribution, viscosity, sedimentation volume ratio, and redispersibility;

[0221] Coating damage is reflected in coating weight gain, coating uniformity, friability, and storage stability.

[0222] A dosage form issue can correspond to multiple key quality attributes, and a key quality attribute can also be affected by multiple formulation parameters and preparation process parameters.

[0223] 2. Establishment of constraint relationships

[0224] The constraints between key quality attributes and formulation parameters and preparation process parameters should include at least the following:

[0225] Allowable parameter range;

[0226] Parameter adjustment direction;

[0227] The direction of the impact of parameter changes on key quality attributes;

[0228] Mutual constraints between different parameters;

[0229] Stability limitations of active ingredients;

[0230] Restrictions on route of administration;

[0231] Bioavailability limitations;

[0232] Drug quality standards are limited.

[0233] For candidate parameter combinations that may alter drug release behavior, preliminary constraints are imposed based on historical in vitro dissolution-in vivo absorption correlations, validated dissolution curve control ranges, or similarity to reference formulation dissolution curves. For candidate parameter combinations whose bioavailability equivalence cannot be confirmed based on existing data, they are not directly deemed to meet the bioavailability constraints, but rather bioequivalence evaluation is included in the validation process.

[0234] Taking conventional release tablets as an example, formulation parameters that affect disintegration and dissolution may include the amount of disintegrant, binder, lubricant, and coating weight gain; preparation process parameters may include granulation endpoint, compression pressure, tablet hardness, coating temperature, and coating speed.

[0235] Increasing the amount of disintegrant generally helps shorten the disintegration time, but excessive disintegrant dosage may affect tablet formability; decreasing the tableting pressure generally helps shorten the disintegration time, but excessively low tableting pressure may lead to unacceptable friability; reducing the coating weight gain may improve early dissolution, but excessively low coating weight gain may affect moisture protection and storage stability. Therefore, candidate parameter combinations should simultaneously satisfy multiple constraints.

[0236] 3. Generation of candidate parameter combinations

[0237] The system generates multiple candidate parameter combinations within the allowable range of formulation and preparation process parameters, and performs the following screening sequentially:

[0238] Eliminate combinations that exceed the permitted scope of the registered prescription;

[0239] Eliminate combinations that may lead to a decrease in the stability of active ingredients;

[0240] Eliminate combinations that are not expected to meet the requirements for disintegration, dissolution, content uniformity, friability, or stability.

[0241] Retain combinations that enable the target's key quality attributes to improve within the target's control range.

[0242] Candidate parameter combinations can be generated using full factorial design, orthogonal design, uniform design, or Latin hypercube sampling. When the number of parameters is small, full factorial design can be used; when the number of parameters is large, experimental design methods can be used to reduce the number of combinations.

[0243] 4. Formulation performance prediction model

[0244] This embodiment uses a quadratic response surface model to predict the disintegration performance, dissolution performance, or release performance corresponding to each candidate parameter combination.

[0245] (1) Model structure

[0246] Regarding the first Key quality attributes are predicted using a quadratic response surface model that includes first-order terms, second-order terms, and parametric interaction terms. The response surface model is as follows:

[0247]

[0248] in, Indicates the first Predicted values ​​of key quality attributes; This represents an input parameter vector consisting of standardized formulation parameters and / or preparation process parameters; Indicates the coefficient of the constant term; Represents the coefficient vector of the linear term; Represents the coefficient matrix corresponding to the quadratic term and the parametric interaction term; superscript Represents the transpose of a vector or matrix.

[0249] (2) Model input and output

[0250] Taking tablets as an example, model inputs can include disintegrant dosage, binder dosage, tableting pressure, and coating weight gain.

[0251] Model outputs can include disintegration time, dissolution rate at 15 minutes, dissolution rate at 30 minutes, tablet hardness, and friability.

[0252] (3) Training data

[0253] The training data comes from the target drug’s completed prescription screening batches, process validation batches, historical qualified production batches, and supplementary test batches prepared for the candidate parameter range.

[0254] This embodiment uses 36 batches with complete prescription parameters, preparation process parameters and test results as modeling data, including 20 prescription research batches, 8 process validation batches and 8 supplementary test batches.

[0255] (4) Data preprocessing

[0256] Input parameters with different dimensions are standardized using range. Batches lacking any core detection result are not used for fitting the corresponding quality attribute model.

[0257] For data that clearly exceeds the process control range, first verify whether it is a recording error; if it is confirmed to be an actual process anomaly, its fitting weight can be retained and reduced so that the model can reflect the performance change trend when the parameters deviate from the normal range.

[0258] (5) Training methods

[0259] The least squares method was used to fit the model coefficients. Quadratic terms or interaction terms with low statistical contribution and no clear relationship with the formulation mechanism were gradually eliminated; interaction terms with a clear relationship with the formulation mechanism were retained, such as the interaction term between disintegrant dosage and tableting pressure.

[0260] Five-fold cross-validation was used to evaluate the predictive performance of the model. If the cross-validation determination coefficient of a key quality attribute is lower than 0.70, or the average relative prediction error is higher than 15%, supplementary experimental data were added and the model was refitted. Candidate parameter combinations were directly selected without using the corresponding model.

[0261] (6) Parameter selection

[0262] Candidate parameter combinations must at least satisfy:

[0263] The predicted collapse timeframe is within the target range;

[0264] Predict that the dissolution rate will reach the target value at a specified time point;

[0265] The predicted hardness and brittleness meet the quality requirements;

[0266] The predicted stability risk does not exceed the preset risk threshold.

[0267] When multiple candidate parameter combinations meet the requirements, the recommended order is determined based on the predicted performance improvement, parameter adjustment range, process feasibility, and confidence level of the corresponding dosage form issues.

[0268] In other embodiments, linear regression models, partial least squares regression models, or other models that can output predicted values ​​of formulation performance may also be used. This invention is not limited to a specific prediction model.

[0269] V. Step S5: Trial production, testing, and correction of constraint relationships

[0270] Candidate parameter combinations that meet the preset requirements for predicted performance improvement and confidence level are selected for trial production. The constraint relationship is corrected based on the deviation between the measured performance and the predicted performance, and the target dosage form, formulation parameter range, preparation process parameter range, and verification items are output.

[0271] The predicted performance improvement is used to characterize the degree to which the predicted performance corresponding to the candidate parameter combination approaches the preset target range relative to the baseline performance corresponding to the current formulation parameters and preparation process parameters. The baseline performance can be the historical stability performance corresponding to the current formulation and preparation process, or the measured performance of the most recent qualified production batch. For key quality attributes such as dissolution rate at a specified time point, where an increase in the value indicates performance improvement, the increase in the candidate predicted value relative to the baseline performance is used as the predicted performance improvement. For key quality attributes such as disintegration time and friability, where a decrease in the value indicates performance improvement, the decrease in the baseline performance relative to the candidate predicted value is used as the predicted performance improvement. For key quality attributes such as tablet hardness and viscosity, which have upper and lower target ranges, the decrease in the distance from the candidate predicted value to the target range relative to the distance from the baseline performance to the target range is used as the predicted performance improvement. When the candidate parameter combination involves multiple key quality attributes, each key quality attribute is compared with its corresponding preset improvement threshold.

[0272] 1. Preparation of trial samples

[0273] Trial production can be carried out through laboratory small-scale tests, pilot-scale tests, or production line verification.

[0274] For tablets, during the trial production process, record the batches of raw and excipient materials, mixing time, granulation endpoint, drying conditions, tableting pressure, tablet hardness, coating temperature, coating speed, and coating weight gain.

[0275] For liquid formulations, record the order of addition of raw materials and excipients, shear rate, homogenization strength, preparation temperature, viscosity adjustment process, and filling conditions.

[0276] 2. Actual performance testing

[0277] Trial samples should be tested according to pharmacopoeia methods, drug registration standard methods, or validated enterprise testing methods. The test items should correspond to the relevant critical quality attributes.

[0278] For example, when the target problem is prolonged onset time, the disintegration time limit and the amount of dissolution at different time points are detected; when the target problem is suspension sedimentation, the particle size distribution, viscosity, sedimentation volume ratio and redispersibility are detected.

[0279] 3. Calculation of prediction bias

[0280] For the For each key quality attribute, the relative deviation between measured and predicted performance is:

[0281]

[0282] in, Indicates relative prediction deviation; Indicates measured performance; Indicates predictive performance; This represents a preset constant used to prevent the denominator from being too small.

[0283] When all core quality attributes reach the preset target range and the deviation between measured performance and predicted performance does not exceed the preset allowable deviation, the recommendation priority of the corresponding candidate parameter combination is increased, and the corresponding constraint relationship is retained.

[0284] The preset allowable deviation can be set from 5% to 15% depending on the detection item.

[0285] When the measured performance fails to reach the preset target range, or when the deviation between the measured performance and the predicted performance exceeds the preset allowable deviation, one or more of the following actions will be performed:

[0286] Adjust the candidate range for the corresponding formulation parameters or preparation process parameters;

[0287] Reduce the confidence level of the corresponding constraint relationship;

[0288] Add the trial production data to the model training data;

[0289] Refit the relationship between parameters and key quality attributes;

[0290] Candidate parameter combinations are regenerated based on the adjusted constraints.

[0291] The above process continues until a candidate parameter combination whose measured performance meets the preset target range is obtained, or the preset number of trials is reached.

[0292] 4. Output of dosage form optimization direction

[0293] The output of dosage form optimization directions should include at least: target dosage form; confirmed dosage form issues; corresponding key quality attributes; formulation parameter range; preparation process parameter range; validation items; confidence level of dosage form issues; predictive performance of candidate parameter combinations and measured performance of trial samples.

[0294] The output results are used to guide subsequent pharmaceutical research and do not directly replace drug change research and registration applications. For changes involving formulation composition, key preparation processes, or release behavior, formulation research, process validation, quality comparison, stability studies, and necessary bioequivalence evaluations should be conducted in accordance with applicable drug change technical requirements.

[0295] IV. Examples

[0296] The following uses conventional release film-coated tablets and dispersible tablets with the same active ingredient as examples to illustrate the implementation process of this invention. To avoid being interpreted as a clinical evaluation targeting a specific marketed drug, the target drug is designated as Drug A.

[0297] Both film-coated tablets and dispersible tablets of drug A contain the same active ingredient, with each unit of dosage containing 250 mg. They have the same indications and routes of oral administration.

[0298] The company found in market feedback data that some users reported prolonged onset time and difficulty swallowing after switching from dispersible tablets to film-coated tablets. However, there were also fluctuations in the dissolution performance of some production batches during the same period. Therefore, it is necessary to determine whether the problems are caused by changes in user condition, inherent reactions of the active ingredient, occasional fluctuations in production batches, or dosage form properties.

[0299] 1. Dataset Composition

[0300] We collected 3,128 market feedback data points over 18 consecutive months, covering 17 production batches of film-coated tablets and dispersible tablets, and obtained 486 dosage form conversion records.

[0301] After verifying the active ingredient, dosage specifications, indications, total daily dosage, and combination therapy regimen, exclude the following records:

[0302] There are 32 records of dosage adjustments made simultaneously during dosage form conversion;

[0303] 21 new records of drugs affecting gastrointestinal status were added during dosage form conversion;

[0304] There were 18 records indicating insufficient observation windows before and after the conversion;

[0305] There are 13 records where user identities cannot be reliably matched.

[0306] A total of 402 user-paired samples were ultimately generated. The relevant data structure is shown in Table 2.

[0307] Table 2. Data Composition of Drug A Example

[0308]

[0309] 2. Extraction of feedback events and negative control events

[0310] Feedback events such as prolonged onset time, difficulty swallowing, stomach discomfort, mild dizziness, fatigue, and tablet surface damage were extracted from market feedback data.

[0311] Among them, prolonged onset time is mainly reflected in disintegration time limit and dissolution amount at specified time points; difficulty swallowing is mainly reflected in tablet size, thickness, surface coating state and disintegration performance; and tablet surface damage is mainly reflected in coating weight gain and friability.

[0312] Mild dizziness and fatigue are systemic reactions related to the active ingredient as described in the drug's package insert, and are recorded in both film-coated tablets and dispersible tablets. The incidence of mild dizziness differed by 1.9 percentage points between the two dosage forms, and the incidence of fatigue differed by 1.3 percentage points, both below the pre-set dosage form difference threshold of 3 percentage points.

[0313] Further calculations revealed that the absolute values ​​of the correlation coefficients between mild dizziness, fatigue, and disintegration time, 15-minute dissolution amount, 30-minute dissolution amount, and coating weight gain were all below 0.15. Therefore, mild dizziness and fatigue were selected as negative control events.

[0314] 3. User-specific issue discrepancies and negative control corrections

[0315] In a paired sample of 402 users, 127 users reported a prolonged onset time after switching to film-coated tablets, while 57 users reported the same problem before the switch.

[0316] After calculation based on occurrence status, frequency of occurrence, and severity, the average in-user problem difference for extended onset time is 0.174.

[0317] In the contemporaneous negative control events, the mean intra-user difference for mild dizziness was 0.021, and the mean intra-user difference for fatigue was 0.015.

[0318] Based on sample size and the reliability of evidence sources, the correction weights for mild dizziness and fatigue were set to 0.55 and 0.45, respectively. The overall difference in negative control events was:

[0319]

[0320] The negative control correction factor is set to 1, resulting in the average corrected difference for prolonged onset time:

[0321]

[0322] The corrected results still indicate that the problem of prolonged onset time worsened after users switched from dispersible tablets to film-coated tablets.

[0323] For the dysphagia problem, the converted mean intra-user problem difference was 0.208, and the corrected problem difference after deducting the difference in negative control events was 0.190.

[0324] 4. Direction of batch performance deviation

[0325] The feedback regarding the extended onset time was correlated with the test data of the corresponding production batches of film-coated tablets. Of the nine film-coated tablet production batches, five batches met the minimum analytical requirements for the number of feedback samples.

[0326] Table 3 shows the differences in disintegration time, 15-minute dissolution amount, and correction issues for each production batch.

[0327] Table 3. Validation of the direction of feedback on extended onset time and production batch test data.

[0328]

[0329] Based on historical qualified production batches, the historical stable range for the disintegration time of film-coated tablets is 8.5 minutes to 12.0 minutes, and the historical stable range for the 15-minute dissolution rate is 75% to 88%.

[0330] Production batches B12, B13, and B15 showed disintegration times exceeding the historical stable range and 15-minute dissolution amounts below the historical stable range. Consequently, the onset time extension issue for these batches significantly increased, thus meeting the pre-defined directional correlation. Production batch B11's disintegration time and 15-minute dissolution amount were both within the historical stable range. Production batch B14 showed no significant change in correction issues, and its corresponding test values ​​were within the historical stable range. Therefore, production batches B11 and B14 were used as stable control batches and were not included in the risk directional consistency rate calculation.

[0331] In the production batches where the three key quality attributes deviated along the risk direction and the correction of the problem differences reached the effective change condition, all three production batches passed the direction verification, and the risk direction consistency rate was 100%.

[0332] Based on the paired samples within 402 user groups, the correction problem difference of 0.1557, the risk direction consistency rate, and the sources of evidence for feedback events, the prolonged onset time of film-coated tablets was determined to be a high-confidence dosage form issue.

[0333] Regarding the swallowing difficulty issue, the tablet diameter and thickness of the corresponding production batches were within the approved range, but the surface roughness of the coating in some production batches was higher than the historical stable level. The proportion of production batches where the swallowing difficulty issue was consistent with the direction of change in surface condition risk was 62%, corresponding to a confidence score of 0.66, thus classified as a medium-confidence dosage form issue.

[0334] 5. Combinations of key quality attributes and candidate parameters

[0335] The onset time was mapped to disintegration time limit and 15-minute dissolution amount, and swallowing difficulty was mapped to tablet thickness, surface coating state, and disintegration performance.

[0336] Based on the existing formulation and production line adjustability of the target drug, the following parameters are selected to generate candidate parameter combinations:

[0337] The amount of croscarmellose sodium used is 2.5% to 4.5%;

[0338] The amount of adhesive used is 1.0% to 2.0%;

[0339] The tablet compression pressure is 70 MPa to 95 MPa;

[0340] The weight gain of the coating is 1.5% to 2.5%.

[0341] A quadratic response surface methodology (RSM) model was established using 36 formulation study batches, process validation batches, and supplementary test batches. The cross-validation coefficients of determination for disintegration time and 15-minute dissolution were 0.86 and 0.82, respectively, with mean relative prediction errors of 7.2% and 8.6%, respectively, meeting the pre-defined model usage conditions.

[0342] The system initially generated 36 candidate parameter combinations. After screening based on stability, friability, hardness, and registered formulation range, 12 candidate parameter combinations were retained, among which 4 candidate parameter combinations achieved the preset improvement threshold for predicted performance.

[0343] The recommended candidate parameter combination with the highest priority is:

[0344] The dosage of croscarmellose sodium is 3.8%;

[0345] Adhesive dosage: 1.4%;

[0346] Tableting pressure 82MPa;

[0347] The coating increased the weight by 1.8%.

[0348] The candidate parameter combination predicts a disintegration time of 8.6 minutes, a dissolution rate of 82.5% at 15 minutes, and a friability of 0.46%.

[0349] 6. Trial production and testing

[0350] Three batches of small-scale samples were prepared according to the above candidate parameter combinations, with each batch having a production scale of 10 kg. The granulation, drying, and mixing processes followed the original preparation process, with adjustments only made to the amount of disintegrant, binder, tableting pressure, and coating weight gain.

[0351] The test results of the trial samples are shown in Table 4.

[0352] Table 4. Trial Production and Testing Results of Candidate Parameter Combinations

[0353]

[0354] The relative deviation between the measured and predicted performance of disintegration time was 5.5%, and the relative deviation of dissolution amount at 15 minutes was 1.5%, both of which were lower than the preset allowable deviation of 10%.

[0355] The disintegration time, dissolution rate at 15 minutes, dissolution rate at 30 minutes, friability, and tablet hardness of the trial samples all met the preset target range.

[0356] Therefore, the constraints corresponding to this candidate parameter combination are retained, its recommendation priority is increased, and the following dosage form optimization directions are output:

[0357] The target dosage form is drug A, a conventional release film-coated tablet;

[0358] The formulation parameters range from 3.5% to 4.1% croscarmellose sodium and 1.2% to 1.6% binder;

[0359] The preparation process parameters range from tableting pressure of 78 MPa to 86 MPa and coating weight gain of 1.6% to 2.0%.

[0360] The validation items include disintegration time, dissolution rate at 15 minutes, dissolution rate at 30 minutes, tablet hardness, friability, content uniformity, and accelerated stability.

[0361] 7. Validation of different processing methods

[0362] To verify the impact of user-matched, negative control event correction, and batch performance deviation direction verification on dosage form problem identification results, 20 candidate feedback problems were selected from 12 production batches. Based on the production batch test results, deviation investigation results, and supplementary test results, 8 of these were identified as dosage form problems that require adjustment of formulation parameters or preparation process parameters.

[0363] Identification was performed using market feedback frequency screening, user-specific matching screening, and three-level attribution processing, and the results are shown in Table 5.

[0364] Table 5. Validation results of different question screening methods

[0365]

[0366] As shown in Table 5, while maintaining the same recognition rate for dosage form issues, intra-user pairing can reduce interference caused by differences in user status; after further negative control event correction and batch performance deviation direction verification, the number of output issues is further reduced, and the consistency with the confirmed results is improved.

[0367] V. System Implementation Method

[0368] The present invention also provides a drug dosage form optimization direction recommendation system based on market feedback data, for implementing the above method.

[0369] The system includes a data association module, a dosage form attribution module, a parameter recommendation module, and a trial production correction module.

[0370] The data association module is used to associate market feedback data for various dosage forms of the target drug, dosage form conversion records of the same user, and test data of the corresponding production batches.

[0371] The dosage form attribution module is used to generate paired samples within users, extract feedback events, perform negative control event correction and batch performance deviation verification, and determine dosage form issues and their confidence levels.

[0372] The parameter recommendation module is used to generate candidate parameter combinations based on dosage form issues confirmed by dosage form attribution, predict the disintegration performance, dissolution performance or release performance corresponding to the candidate parameter combinations, and output the dosage form optimization direction.

[0373] The trial production correction module is used to receive the measured performance of the trial production sample, determine the deviation between the measured performance and the predicted performance, and correct the constraint relationship between key quality attributes, formulation parameters, and preparation process parameters based on the deviation.

[0374] The modules described above can be deployed on the same server, or separately on the pharmacovigilance platform, quality management platform, and formulation development platform, exchanging structured data through data interfaces. This module division is for illustrative purposes only and does not limit the actual software deployment method or hardware architecture.

[0375] VI. Implementation Methods of Electronic Equipment

[0376] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor.

[0377] The memory is used to store the computer program, as well as market feedback data, dosage form conversion records, production batch test data, the correspondence between feedback issues and key quality attributes, the constraint relationship between key quality attributes and formulation parameters and preparation process parameters, candidate parameter combinations, and the measured performance of trial samples.

[0378] When the processor executes the computer program, it performs the following processes:

[0379] Obtain market feedback data for each dosage form of the target drug, dosage form conversion records for the same user, and corresponding production batch testing data, and correlate them;

[0380] Create user-paired samples, extract feedback events, select negative control events, and determine batch performance deviations;

[0381] Identify the differences in user-specific issues and the differences in concurrent negative control events, obtain the corrected issues, and perform batch performance deviation direction verification;

[0382] The dosage form issues identified through dosage form attribution are mapped to key quality attributes, and candidate parameter combinations are generated and screened.

[0383] The constraint relationship is corrected based on the measured performance of the trial sample, and the direction for dosage form optimization is output.

[0384] The electronic device can be a server for a pharmaceutical manufacturing company, a formulation research and development workstation, a quality data processing platform, a cloud computing server, or other computing devices with data storage and processing capabilities.

[0385] The processor may be a central processing unit, a digital signal processor, or other processing device capable of executing computer programs. The memory may include volatile memory and non-volatile memory, used to store program instructions and data required to execute the program instructions.

[0386] During operation, the electronic device can connect to the market feedback database, production batch testing database, and formulation R&D database via data interfaces, or store the relevant data locally on the electronic device. The processor initiates the corresponding processing flow according to a preset execution cycle or upon receiving new market feedback data, production batch testing data, or trial sample testing results.

Claims

1. A method for recommending drug dosage form optimization directions based on market feedback data, characterized in that, include: S1. Obtain one or more of the following: market feedback data for each dosage form of the target drug, dosage form conversion records for the same user, and disintegration, dissolution, particle size, viscosity, and stability test data for the corresponding production batch and the dosage form; and associate them according to user, active ingredient, dosage specification, indication, production batch, and stage of administration. S2. Form a paired sample within the user based on market feedback data before and after the same user's dosage form conversion and extract feedback events; select negative control events from systemic reactions that are present in all dosage forms and caused by the pharmacological effects of the active ingredients; determine batch performance deviations based on the deviations of the test data from the registration standard control range or historical stable intervals; S3. For each feedback issue, determine the intra-user issue difference, subtract the difference of the negative control event in the same period, and obtain the corrected issue difference; when the batch performance deviation and the corrected issue difference meet the preset direction correspondence, confirm the dosage form attribution and obtain the dosage form issue and confidence level; S4. Map the confirmed dosage form issues to key quality attributes. Based on the constraints between key quality attributes and formulation parameters and preparation process parameters, generate candidate parameter combinations that meet the constraints of active ingredient stability, route of administration, bioavailability and drug quality standards, and predict the disintegration, dissolution, release, particle size, viscosity, dispersion or stability performance corresponding to each candidate parameter combination. S5. Select candidate parameter combinations that meet the preset requirements for predicted performance improvement and confidence level, and conduct trial production. Correct the constraint relationship based on the deviation between the measured performance and the predicted performance, and output the target dosage form, prescription parameter range, preparation process parameter range, and verification items.

2. The method for recommending drug dosage form optimization directions based on market feedback data according to claim 1, characterized in that, In step S1: The market feedback data includes one or more of the following: drug evaluation texts, complaint records, records of reasons for drug return or exchange, medication adherence records, adverse reaction reports, and follow-up records. The dosage form conversion record includes at least the user identifier, dosage form before conversion, dosage form after conversion, conversion time, dosage, and production batch identifier; The market feedback data is associated with the corresponding production batch test data based on the production batch identifier, and the association is verified based on the generation time, dosage form conversion time, and medication stage of the market feedback data.

3. The method for recommending drug dosage form optimization directions based on market feedback data according to claim 1, characterized in that, In step S2, forming the user-in-user pairing sample includes: Filter market feedback data generated by the same user using the first and second dosage forms of the target drug within a preset pairing time range; Verify that the active ingredients, dosage specifications, indications, and dosing regimens corresponding to the first and second dosage forms are consistent; Exclude dosage form conversion records that involve changes in active ingredients, dosage specifications, indications, or combination therapy regimens during the dosage form conversion period; The market feedback data before and after the dosage form conversion that meet the verification criteria will be used to form a set of paired samples within users. And / or, extracting the feedback event includes: Standardize the use of synonyms, negations, degree expressions, and time expressions in market feedback data; Extract feedback events from standardized market feedback data, which consist of the problem type, stage of occurrence, severity, source of evidence, and dosage form attributes involved. Duplicate feedback events are merged based on user identifier, medication stage, time of occurrence of feedback event, and content of feedback event; The dosage form attributes include one or more of the following: dosage form size, surface coating, disintegration properties, dissolution properties, release properties, particle state, viscosity state, dispersion state, and storage stability.

4. The method for recommending drug dosage form optimization directions based on market feedback data according to claim 1, characterized in that, In step S2, the negative control event is selected according to the following conditions: The negative control event is a systemic reaction related to the active ingredient recorded in the drug instructions or a pre-set pharmacological knowledge base; The negative control events are recorded in each of the comparison dosage forms of the target drug; The difference in the occurrence of the negative control event among the various comparison dosage forms is lower than a preset dosage form difference threshold. The correlation between the negative control event and the data on disintegration, dissolution, particle size, viscosity, and stability is lower than a preset correlation threshold; And / or, determining the batch performance deviation includes: The disintegration, dissolution, particle size, viscosity, and stability test values ​​of each production batch were compared with the corresponding registration standard control ranges. When the registration standard does not specify the corresponding control range, the test value is compared with the historical stable range determined based on the historical qualified batches of the target drug; Based on the deviation index, direction of deviation, and degree of deviation of the detected value relative to the registered standard control range or historical stable interval, a batch performance deviation record for the corresponding production batch is generated.

5. The method for recommending drug dosage form optimization directions based on market feedback data according to claim 1, characterized in that, In step S3: The differences in user-related issues are determined based on the occurrence status, frequency, and severity of feedback events before and after the dosage form change; The differences in negative control events during the same treatment period are determined based on the occurrence status, frequency, and severity of the negative control events. The corrected problem difference is obtained by subtracting the impact of the difference in the negative control event during the same period from the difference in the problem within the user; When multiple negative control events exist, the correction priority of each negative control event is determined based on the sample size, record completeness, and reliability of the evidence source.

6. The method for recommending drug dosage form optimization directions based on market feedback data according to claim 1, characterized in that, In step S3: The preset directional correspondence is a pre-established correspondence between feedback issues, key quality attributes, and the risk change direction of key quality attribute detection values; When the difference in the correction problem indicates that the corresponding dosage form problem has worsened, and the detection value of the corresponding key quality attribute changes in the risk direction of deviating from the target control range, it is determined that the batch performance deviation and the difference in the correction problem satisfy a preset directional correspondence. When the difference in the correction problem indicates that the corresponding dosage form problem has been alleviated, and the detection value of the corresponding key quality attribute returns to the target control range, it is determined that the batch performance deviation and the difference in the correction problem satisfy a preset directional correspondence. If any of the above conditions are not met, it is not confirmed that the corresponding feedback event was caused by the dosage form attribute. And / or, the confidence level of the dosage form problem is determined based on the number of user-paired samples that form the difference in the correction problem, the degree of difference in the correction problem, the number and proportion of production batches that meet the preset directional correspondence, and the reliability of the evidence source of the feedback event; According to the preset confidence classification conditions, the confidence level is divided into high confidence level, medium confidence level or low confidence level.

7. The method for recommending drug dosage form optimization directions based on market feedback data according to claim 1, characterized in that, In step S4: Step S4 generates candidate parameter combinations only for dosage form issues that reach a medium or high confidence level; The key quality attributes include one or more of the following: disintegration time, dissolution profile, release rate, particle size distribution, viscosity, dispersion uniformity, and storage stability. The formulation parameters include one or more of the following: type and dosage of disintegrant, type and dosage of binder, coating material and coating weight gain, type and dosage of suspending agent and type and dosage of release modifier. The preparation process parameters include one or more of the following: granulation endpoint, tableting pressure, coating temperature, coating speed, particle size, homogenization strength, and drying conditions. The constraint relationship is used to characterize the relationship between the adjustment direction and range of the prescription parameters or preparation process parameters and the change direction of the corresponding key quality attributes.

8. The method for recommending drug dosage form optimization directions based on market feedback data according to claim 1, characterized in that, In step S5: The predicted performance improvement is the degree to which the predicted performance corresponding to the candidate parameter combination approaches the preset target range relative to the baseline performance. The baseline performance is the historical stability performance or the measured performance of the most recent qualified production batch corresponding to the current formulation parameters and preparation process parameters. When the candidate parameter combination corresponds to multiple key quality attributes, the improvement in the predictive performance of each key quality attribute is compared with the corresponding preset improvement threshold. The prototype samples were prepared according to the candidate parameter combinations, and the prototype samples were subjected to performance tests corresponding to the key quality attributes. When the measured performance reaches the preset target range, the recommendation priority of the corresponding candidate parameter combination is increased, and the corresponding constraint relationship is retained; When the measured performance does not reach the preset target range, or the deviation between the measured performance and the predicted performance exceeds the preset allowable deviation, the candidate range of the corresponding formulation parameters or preparation process parameters is adjusted, and the confidence level of the corresponding constraint relationship is reduced. Based on the adjusted constraint relationships, candidate parameter combinations are regenerated until a candidate parameter combination whose measured performance meets the preset target range is obtained, or the preset number of trial productions is reached.

9. A drug dosage form optimization direction recommendation system based on market feedback data, characterized in that, The system is used to implement the drug dosage form optimization direction recommendation method based on market feedback data according to any one of claims 1 to 8, the system comprising: The data association module is used to associate market feedback data for various dosage forms of the target drug, dosage form conversion records of the same user, and test data of the corresponding production batches. The dosage form attribution module is used to generate in-user paired samples, extract feedback events, perform negative control event correction and batch performance deviation verification, and determine dosage form issues and their confidence levels. The parameter recommendation module is used to generate candidate parameter combinations based on dosage form issues confirmed by dosage form attribution, predict the formulation performance corresponding to the candidate parameter combinations, and output the dosage form optimization direction. The trial production correction module is used to correct the constraint relationship between key quality attributes and formulation parameters and preparation process parameters based on the measured performance of the trial production sample.

10. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, it implements the method for recommending drug dosage form optimization directions based on market feedback data as described in any one of claims 1 to 8.

Images