Antiepileptic Patient Medication Assessment Management System Based on Chromatography-Combined Data

Abstract

This invention relates to the field of biomedical testing and clinical medication management technology, specifically to an antiepileptic patient medication assessment and management system based on chromatography-mass spectrometry (CMS) data. The system includes: a data acquisition module for collecting multidimensional time-series data generated by CMS in patient biosamples, and obtaining prescription dosage, administration timestamps, weight, and genotyping data; a benchmark reconstruction module for constructing an ideal pharmacokinetic benchmark based on a compartmental model and the Michaelis-Menten kinetic rules; a parameter injection module for generating a theoretically abnormal simulation state; a differential extraction module for generating a real residual vector and a theoretical residual matrix; a coupling decision module for outputting the state assessment results; and a management feedback module for generating medication reminders, dosage adjustment suggestions, medication change prompts, resampling, or manual review instructions, achieving a closed-loop connection from testing data to clinical decision-making.

Description

Technical Field

[0001] This invention relates to the field of biomedical testing and clinical drug management technology, specifically to an antiepileptic patient medication evaluation and management system based on chromatography-chromatographic combined data. Background Technology

[0002] With the development of antiepileptic drug therapy and chromatography-mass spectrometry (GC-MS) detection technology, the assessment and management of patients' long-term medication process has become an important part of clinical follow-up. In order to effectively identify patients' medication status and make subsequent interventions, it is particularly important to conduct joint analysis of test data and individual information.

[0003] In existing antiepileptic drug monitoring, blood drug concentration determination largely relies on single sampling results and general reference ranges. While these can reflect baseline concentration levels, for patients undergoing long-term treatment, fluctuations in adherence, differences in metabolic capacity, and sampling interference can all lead to similar test results. Especially in outpatient follow-up scenarios, relying solely on a single concentration value makes it difficult to effectively utilize the continuous changes in the parent drug and metabolites, and it is also difficult to accurately distinguish the source of abnormalities by combining information such as dosing time, prescription dosage, weight, and genotyping. Therefore, it is not conducive to the precise formulation of subsequent management measures.

[0004] Therefore, how to process existing chromatographic-mass spectrometry detection data and individual patient medication-related data to obtain corresponding medication status assessment results is crucial for establishing a more reliable anomaly identification mechanism based on pharmacokinetic principles, and for forming corresponding reminder, medication adjustment, medication change, resampling, or manual verification strategies. This is essential for ensuring the safety and effectiveness of antiepileptic medication. Summary of the Invention

[0005] The purpose of this invention is to provide a medication evaluation and management system for antiepileptic patients based on chromatography-chromatographic combined data, and to solve the following technical problems:

[0006] This avoids relying solely on single concentration values ​​or high-low comparisons, which can mask individual metabolic differences or confuse genuine pharmaceutical abnormalities with instrument detection interference. It also makes it easier to distinguish the true triggers of detection deviations in a finer granular way, such as abnormal compliance, metabolic abnormalities, and data collection interference. This enables a closed-loop connection from multidimensional detection data to the generation of specific business management instructions.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] An antiepileptic patient medication assessment and management system based on chromatography-chromatographic data includes:

[0009] The data acquisition module is used to collect multidimensional time-series data generated by chromatography-mass spectrometry detection of the parent antiepileptic drug and its metabolites in the patient's biological samples, and to obtain the patient's prescription dosage, administration timestamp, weight and genotyping data from the electronic medical record system;

[0010] The baseline reconstruction module is used to construct an ideal pharmacokinetic baseline based on the prescribed dosage, administration time stamp, body weight, and genotyping data, using a compartmental model and the Michaelis-Menten kinetic rule, and to generate reference concentration time series and ratio curves of the parent drug and metabolites.

[0011] The parameter injection module is used to perform perturbation simulation on the ideal pharmacokinetic benchmark by using preset compliance abnormal parameters, preset metabolic abnormal parameters and preset acquisition interference parameters obtained from the system preset abnormal template library, respectively, so as to generate the corresponding theoretical abnormal simulation state.

[0012] The difference extraction module is used to generate a real residual vector based on the difference between the multidimensional time series data and the ideal pharmacokinetic benchmark, and to subtract each of the theoretical abnormal simulation states from the ideal pharmacokinetic benchmark to obtain multiple theoretical residual vectors, and to combine the multiple theoretical residual vectors to form a theoretical residual matrix.

[0013] The coupling decision module is used to calculate the similarity between the actual residual vector and each of the theoretical residual vectors in the theoretical residual matrix, and output the state evaluation result based on the calculation result. The state evaluation result includes at least one of the following: compliance abnormal state, metabolic abnormal state, acquisition interference state, mixed abnormal state, and state awaiting manual review.

[0014] The management feedback module is used to generate medication reminder instructions, dosage adjustment suggestion instructions, medication change reminder instructions, resampling instructions, or manual review instructions based on the status assessment results, and send the instructions to the corresponding medical workstation or patient interaction terminal for feedback execution.

[0015] Optionally, the multidimensional time-series data includes parent drug data and metabolite data; both the parent drug data and the metabolite data include time-series data formed by the changes in retention time, mass-to-charge ratio, and peak area integral over continuous sampling time.

[0016] Optionally, the baseline reconstruction module is used to perform time-progression solving on the equations describing pharmacokinetics to generate the ideal pharmacokinetic baseline; wherein, when constructing the ideal pharmacokinetic baseline, it is set to use a baseline metabolic parameter state that matches the genotyping data, and is set to an ideal physical detection state without introducing the preset acquisition interference parameters.

[0017] Optionally, the theoretical anomaly simulation states include compliance anomaly simulation states, metabolic anomaly simulation states, and disturbance anomaly simulation states, and the parameter injection module includes:

[0018] The irregular injection unit is used to introduce the preset compliance anomaly parameters into the ideal pharmacokinetic benchmark, wherein the preset compliance anomaly parameters include missed dose parameters and / or incorrect dose parameters, so as to generate the compliance anomaly simulation state.

[0019] A metabolic infusion unit is used to introduce the preset metabolic abnormality parameters into the ideal pharmacokinetic baseline, wherein the preset metabolic abnormality parameters include enzyme induction parameters and / or enzyme inhibition parameters to generate the metabolic abnormality simulation state.

[0020] An interference injection unit is used to introduce the preset acquisition interference parameters into the ideal pharmacokinetic baseline, wherein the preset acquisition interference parameters include ion suppression parameters to generate the interference anomaly simulation state.

[0021] Optionally, the difference extraction module is used to first perform time-keeping alignment, peak area normalization, and filtering processing to suppress baseline drift and high-frequency noise on the multidimensional time-series data in sequence, and then generate the actual residual vector.

[0022] Optionally, the coupling decision module is used to calculate the dynamic time warp distance and Mahalanobis distance between the actual residual vector and each of the theoretical residual vectors in the theoretical residual matrix, and to convert the dynamic time warp distance and the Mahalanobis distance into classification similarity by calculating the reciprocal of the distance value or applying a negative exponential function, thereby generating classification similarity values ​​corresponding to the compliance abnormality state, the metabolic abnormality state and the acquisition interference state, respectively.

[0023] Optionally, the decision rule configuration of the coupled decision module is as follows: determine whether the generated classification similarity values ​​are all less than a preset decision threshold, wherein the decision threshold and the preset proximity threshold are pre-configured and calibrated based on the consistency rate between automatic classification and manual decision conclusions in historical review samples;

[0024] If all the classification similarity values ​​are less than the decision threshold, the state evaluation result is determined and output as the state to be manually reviewed.

[0025] If there is a classification similarity value that is not less than the decision threshold, then calculate the difference between the extracted maximum classification similarity and the second maximum classification similarity.

[0026] When the difference is not greater than a preset proximity threshold or there are ties for the maximum classification similarity, the state evaluation result is determined and output as the mixed abnormal state.

[0027] When the difference is greater than the proximity threshold and the maximum classification similarity is unique, the state subdivision judgment is performed based on the abnormal state category corresponding to the maximum classification similarity as the initial abnormal category: when the initial abnormal category is the compliance abnormal state, the corresponding missing or incorrect dosage sub-state of the compliance abnormal state is output; when the initial abnormal category is the metabolic abnormal state, the corresponding metabolic abnormal state is output; when the initial abnormal category is the acquisition interference state, the corresponding interference verification sub-state of the acquisition interference state is output.

[0028] Optionally, the management feedback module is specifically used for:

[0029] In response to the status assessment result being the missed or incorrect dose sub-status, the medication reminder instruction is generated;

[0030] In response to the status assessment result being the metabolic abnormality, a dose adjustment suggestion instruction or a medication change prompt instruction is generated for medical staff to review;

[0031] In response to the state assessment result being the interference complex sub-state, the resampling instruction is generated;

[0032] In response to the status assessment result being either the status awaiting manual review or the mixed abnormal status, the manual review instruction is generated.

[0033] Optionally, the management feedback module is also used to receive external review input results for manual review instructions, and update the individual characteristic metabolic parameters in the ideal pharmacokinetic benchmark based on the external review input results, and synchronously and adaptively correct the decision threshold and the proximity threshold, so as to provide a parameter determination basis for iterative optimization for subsequent system status assessment.

[0034] The beneficial effects of this invention are:

[0035] 1. This invention constructs an ideal pharmacokinetic baseline by combining patient prescription dosage, administration time, weight, and genotyping data, overcoming the limitations of traditional single sampling and general intervals; this individualized baseline truly reflects the normal medication status under the patient's inherent physiological background, effectively avoiding false alarms and missed alarms caused by individual metabolic differences, and laying the foundation for subsequent accurate anomaly identification;

[0036] 2. This invention establishes a multidimensional time-series data organization and real residual preprocessing mechanism; by synchronously processing retention time, mass-to-charge ratio and peak area integral, and performing time alignment, normalization and filtering noise reduction before differencing, technical noise such as instrument drift and matrix interference is effectively eliminated, so that the residual vector has higher data quality and clinical sensitivity.

[0037] 3. This invention introduces a parameter injection and theoretical residual matrix generation mechanism, combined with dynamic time warping and Mahalanobis distance dual algorithms, which effectively solves the problem of distinguishing highly similar abnormal appearances; the system not only considers the temporal pattern, but also takes into account the joint distribution law of multiple variables, and can accurately attribute a single low concentration to a real cause such as missed dose, wrong dose, enzyme induction or ion inhibition.

[0038] 4. This invention designs a multi-level state judgment and business linkage mechanism for threshold triage, which solves the technical problem of the disconnect between simple data analysis and actual clinical intervention; by accurately distinguishing between mixed abnormalities and single abnormalities, the system automatically distributes medication reminders, dosage adjustment suggestions, resampling or manual review instructions, realizing efficient integrated linkage between outpatient follow-up, clinical pharmacy and laboratory department;

[0039] 5. This invention constructs a feedback optimization mechanism for manual review instructions, overcoming the shortcomings of static models that cannot adapt to long-term metabolic changes in patients; by receiving external medical and nursing review results, the system dynamically corrects the individual characteristic metabolic parameters and judgment thresholds in the benchmark, enabling the system to continuously self-calibrate and iteratively evolve during long-term follow-up. Attached Figure Description

[0040] The invention will now be further described with reference to the accompanying drawings.

[0041] Figure 1 This is a schematic diagram of the modules of the antiepileptic patient medication evaluation and management system based on chromatography-chromatographic data provided in the embodiments of this application. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Please see Figure 1 The antiepileptic patient medication assessment and management system based on chromatography-mass spectrometry data includes: a data acquisition module, used to acquire multidimensional time-series data generated by chromatography-mass spectrometry detection of the parent antiepileptic drug and its metabolites in the patient's biological samples, and to obtain the patient's prescription dosage, administration timestamp, weight and genotyping data from the electronic medical record system;

[0044] The baseline reconstruction module is used to construct an ideal pharmacokinetic baseline based on the prescribed dosage, administration time stamp, body weight, and genotyping data, using a compartmental model and the Michaelis-Menten kinetic rule, and to generate reference concentration time series and ratio curves of the parent drug and metabolites.

[0045] The parameter injection module is used to perform perturbation simulation on the ideal pharmacokinetic benchmark by using preset compliance abnormal parameters, preset metabolic abnormal parameters and preset acquisition interference parameters obtained from the system preset abnormal template library, respectively, so as to generate the corresponding theoretical abnormal simulation state.

[0046] The difference extraction module is used to generate a real residual vector based on the difference between the multidimensional time series data and the ideal pharmacokinetic benchmark, and to subtract each of the theoretical abnormal simulation states from the ideal pharmacokinetic benchmark to obtain multiple theoretical residual vectors, and to combine the multiple theoretical residual vectors to form a theoretical residual matrix.

[0047] The coupling decision module is used to calculate the similarity between the actual residual vector and each of the theoretical residual vectors in the theoretical residual matrix, and output the state evaluation result based on the calculation result. The state evaluation result includes at least one of the following: compliance abnormal state, metabolic abnormal state, acquisition interference state, mixed abnormal state, and state awaiting manual review.

[0048] The management feedback module is used to generate medication reminder instructions, dosage adjustment suggestion instructions, medication change reminder instructions, resampling instructions, or manual review instructions based on the status assessment results, and send the instructions to the corresponding medical workstation or patient interaction terminal for feedback execution.

[0049] This embodiment provides a medication assessment and management mechanism for antiepileptic patients based on chromatography-chromatographic combined data; specifically, the system is deployed at a joint workstation between the hospital's epilepsy clinic and laboratory department to conduct continuous medication assessment for follow-up patients receiving long-term antiepileptic treatment;

[0050] A typical application scenario is set as follows: Patient A receives maintenance treatment with antiepileptic drugs such as valproic acid or carbamazepine in the outpatient clinic due to recurrent seizures. The doctor prescribes a fixed dosage regimen. The laboratory collects blood samples from the patient at multiple follow-up points and completes chromatography-mass spectrometry testing. The system provides medication status assessment and subsequent management instructions at the same time as the test results are issued.

[0051] Specifically, the data acquisition module receives the data file output by the detection instrument and extracts the parent drug peak, major metabolite peak, retention time, mass-to-charge ratio, and peak area integral value corresponding to each sampling time point. At the same time, it reads the prescription dose, the most recent administration time stamp, the patient's weight, and the existing genotyping results from the electronic medical record system.

[0052] For ease of technical explanation, the following calculation parameter extrapolation example can be used: If a patient takes medication consecutively at 08:00 on day 1 and 08:00 on day 2, with a dose of 200mg each time, weighs 50kg, and genotyping indicates a common metabolite, then the system will obtain a set of actual test values ​​when blood is drawn at 10:00 on day 2, for example, the peak area of ​​the parent drug is 82 and the peak area of ​​the metabolite is 35; if another blood is drawn at 14:00 on day 2, the peak area of ​​the parent drug is 60 and the peak area of ​​the metabolite is 41; these two sampling values, together with the retention time and mass-to-charge ratio information, constitute a set of multidimensional time-series inputs;

[0053] Furthermore, the baseline reconstruction module does not directly adopt a general reference interval, but generates an ideal pharmacokinetic baseline based on individual patient parameters. This baseline can be understood as the time trajectory of the parent drug and metabolites that should appear when the patient takes medication as prescribed, has metabolic capacity at the baseline state corresponding to their genotype, and the detection process is free from acquisition interference. In practical applications, the body can be regarded as one or more compartments. The system matches the preset compartment structure based on the physicochemical properties of the currently prescribed drugs read from the electronic medical record and the pharmacokinetic guidelines. For example, a single-compartment model is used by default for drugs with small distribution volumes such as valproic acid, while a two-compartment model is used for drugs with high lipid solubility and large distribution volumes. After calling the intercompartment clearance parameter corresponding to the model, the concentration change is gradually advanced using the time step.

[0054] At each time step, the concentration of the parent drug is updated according to the absorption and clearance rules, and the concentration of metabolites is updated according to the formation and elimination rules. When there is a nonlinear metabolic saturation characteristic, the Michaelis-Menten rule is introduced to correct the elimination rate.

[0055] Taking the above patient as an example, the system calculated that at 10:00, the peak area corresponding to the ideal parent drug concentration should be about 90, and the metabolite should be about 33; at 14:00, the ideal parent drug concentration should be about 63, and the metabolite should be about 40, thus forming a parent drug reference time series, a metabolite reference time series, and a parent drug / metabolite ratio curve.

[0056] With only an ideal benchmark, the system can only identify the degree of numerical deviation in a single dimension based on the comparison between high and low, and it is difficult to distinguish the real triggering source of the deviation. To this end, the parameter injection module actively applies multiple types of perturbations to the ideal benchmark to form several theoretically abnormal simulation states.

[0057] First, compliance abnormal parameters can simulate missed doses, delayed doses, or incorrect doses; second, metabolic abnormal parameters can simulate changes in metabolic rate caused by liver enzyme induction or inhibition; third, acquisition interference parameters can simulate ion inhibition, local peak distortion, or occasional detection shifts.

[0058] For example, in the missed dose simulation, the theoretical value of the parent drug at 10:00 may decrease from 90 to 50, while the metabolite only decreases from 33 to 28 due to the lag effect; in the enzyme-induced simulation, the theoretical value of the parent drug at 10:00 may be 58, while the metabolite increases to 46; in the acquisition interference simulation, the overall concentration trend of the parent drug and metabolite remains approximately unchanged, but the peak area at a certain point in time is disproportionately reduced, such as the parent drug being perturbed from 90 to 72 and the metabolite from 33 to 25, while high-frequency noise is superimposed near the retention time window;

[0059] The differential extraction module generates two types of residuals; the actual residual vector is obtained by subtracting the ideal baseline from the actual sampled data. For example, at 10:00, the residual of the parent drug is 82-90=-8, and the residual of the metabolite is 35-33=2; at 14:00, the residual of the parent drug is 60-63=-3, and the residual of the metabolite is 41-40=1.

[0060] By concatenating these in chronological order, we can obtain the actual residual vector. Correspondingly, after subtracting the ideal baseline from each theoretical anomaly simulation state, multiple theoretical residual vectors are obtained, such as the missed iteration residual. Enzyme-induced residual Collect interference residuals Multiple theoretical residual vectors are then combined into a theoretical residual matrix; each column or row of this matrix corresponds to an anomaly template, which facilitates subsequent unified matching.

[0061] The coupling decision module calculates the similarity between the actual residual vector and each template in the theoretical residual matrix. To illustrate the logic, a simplified similarity value can be used first: if the similarity between the actual residual and the missed dose template is 0.42, the similarity with the enzyme-induced template is 0.80, and the similarity with the acquisition interference template is 0.36, then the system outputs a metabolic abnormality status.

[0062] If the values ​​of the missed dose template and the enzyme-induced template reach 0.73 and 0.71 respectively, the system will output a mixed abnormal status because the two are very close; the management feedback module will generate specific business instructions based on the output results.

[0063] For example, medication reminders are sent to patients for missed doses, dosage adjustment suggestions or medication change prompts are sent to medical staff for metabolic abnormalities, resampling tasks are sent to the laboratory for cases of data collection interference, and cases that cannot be clearly classified are entered into a manual review process.

[0064] As a backup fault tolerance mechanism, when input data is missing, dosing timestamps conflict, or genotyping has not yet been returned, the system can first mark it as an incomplete baseline and pause the automatic dosing recommendation, only retaining the resampling or manual review exit; when the number of consecutive sampling points does not reach the preset lower limit, for example, when there is only one isolated sampling point, the system only performs initial screening and does not output detailed classification results of missed doses and metabolic abnormalities.

[0065] If neither the parent drug peak nor the metabolite peak in the detection file reaches the minimum signal-to-noise ratio (SNR), the minimum SNR refers to the lower limit of the ratio of the signal intensity of the target analyte chromatographic peak to the baseline noise intensity. It is pre-calibrated based on the quantitation limit test results of historical healthy population blank sample matrix superimposed with standards, and is usually set to SNR S / N=10. The current batch is directly determined to be unusable for modeling, and a sampling failure prompt is triggered.

[0066] On the day of the outpatient follow-up visit, Patient A stated that he had been taking his medication on time recently. However, the actual residuals obtained from the two samplings by the system were closer to the theoretical template of enzyme induction and significantly deviated from the template of missed doses. As a result, the medical staff received a reminder to review the metabolic abnormality. The doctor can further determine whether there is an effect of combined use of enzyme-inducing drugs by combining clinical episodes and medications. The purpose of this step is to subdivide the low concentration of a single test into manageable cause categories, thereby achieving a closed-loop connection from test data to clinical decision-making.

[0067] In a preferred embodiment of the present invention, the multidimensional time-series data includes parent drug data and metabolite data; both the parent drug data and the metabolite data include time-series data formed by the changes in retention time, mass-to-charge ratio, and peak area integral with continuous sampling time.

[0068] This embodiment provides a multidimensional time-series data organization mechanism; specifically, in the aforementioned outpatient follow-up scenario, the system does not only save the single value of blood drug concentration from a single blood draw, but simultaneously saves the retention time, mass-to-charge ratio, and peak area integral of the parent drug and metabolites at multiple consecutive sampling points, so as to identify the source of anomalies from three levels: chromatographic behavior, mass spectrometry behavior, and pharmacokinetics behavior.

[0069] Specifically, at a single sampling point, the parent drug data includes at least one target retention time window, the mass-to-charge ratio of the corresponding target ion pair, and the integrated area of ​​the ion peak; the metabolite data also adopts the same structure.

[0070] For example, if patient A is sampled at two time points, 10:00 and 14:00, the system can generate the following simplified data fragments: at 10:00, the parent drug has a retention time of 5.2 min, a mass-to-charge ratio of 301, and a peak area of ​​82, while the metabolite has a retention time of 4.8 min, a mass-to-charge ratio of 257, and a peak area of ​​35; at 14:00, the parent drug has a retention time of 5.2 min, a mass-to-charge ratio of 301, and a peak area of ​​60, while the metabolite has a retention time of 4.8 min, a mass-to-charge ratio of 257, and a peak area of ​​41.

[0071] In this way, the system can subsequently compare the numerical fluctuations of the peak area over time, and also observe whether the retention time drifts overall and whether the mass-to-charge ratio shows abnormal jumps.

[0072] The reason for introducing this multidimensional structure is that if we rely solely on the peak area sequence, it is easy to confuse the actual metabolic increase with sampling interference. For example, if the peak area of ​​the parent drug in a certain sample is low, it may be due to the patient missing a dose, or it may be due to the chromatographic retention time shift causing incomplete integration window truncation, or it may be due to ion inhibition causing low response pressure.

[0073] After incorporating retention time and mass-to-charge ratio, the system can further determine: if the peak area decreases but the retention time and mass-to-charge ratio remain stable, it is more likely to be a pharmacokinetic or compliance issue; if the peak area decreases while the peak shape is significantly skewed within the retention time window, it is more likely to be interference on the detection side.

[0074] As a backup fault tolerance mechanism, when only the parent drug peak is identified at a certain time point but no metabolite peak is identified, the system does not directly delete that time point, but fills it into the sequence with a missing detection marker and reduces the weight of that time point in subsequent matching; if the retention time offset exceeds the preset tolerance range, but the mass-to-charge ratio and peak shape can still confirm the target compound, the system can perform alignment correction before adding it to the database; if the mass-to-charge ratio deviates too much and cannot match the target compound, then that time point is invalidated and triggers reanalysis or resampling;

[0075] In the same outpatient follow-up process, a sample from Patient A showed that the peak area of ​​the parent drug decreased from 82 to 54, but its retention time drifted from 5.2 min to 5.8 min, and the peak shape was truncated. Based on this, the system will not immediately push a missed dose reminder, but will prioritize entering the collection interference screening link. The purpose of this step is to make the subsequent judgment based on the complete detection behavior characteristics, so as to effectively distinguish between pharmaceutical abnormalities and instrument abnormalities.

[0076] In a preferred embodiment of the present invention, the baseline reconstruction module is used to perform time-progression solving of the equations describing pharmacokinetics to generate the ideal pharmacokinetics baseline; wherein, when constructing the ideal pharmacokinetics baseline, it is set to use a baseline metabolic parameter state that matches the genotyping data, and is set to an ideal physical detection state without introducing the preset acquisition interference parameters.

[0077] This embodiment provides a time-progressive reconstruction mechanism for an ideal pharmacokinetic baseline. Specifically, in the aforementioned continuous follow-up scenario, relying solely on the population mean reference line can easily mask individual differences. Especially in the treatment of antiepileptic drugs, patients with different metabolites may exhibit significantly different concentration curves even if they take the medication exactly as prescribed. Therefore, this embodiment introduces an individualized baseline reconstruction process that is solved step by step according to the time step.

[0078] Specifically, the system first determines the discrete time axis according to the dosing regimen, for example, taking 30 minutes as a step, progressing from several hours before the most recent dosing to the current sampling time; in each step, the parent drug amount is updated according to the absorption and elimination amount, and the metabolite amount is updated according to the generation and clearance amount;

[0079] If the drug is eliminated approximately linearly within the normal concentration range, the rate of elimination can be reduced by a fixed percentage; if it enters the enzyme saturation region or approaches the upper limit of metabolism, the Michaelis-Menten rule is used to control the elimination rate so that it does not increase indefinitely.

[0080] Specifically, at each time step If the concentration of the parent drug is within the range... Below a preset saturation threshold, it eliminates concentration changes. Based on the first-order clearance rate constant Perform the calculation:

[0081]

[0082] If concentration If the preset saturation threshold is exceeded, then Switch to based on maximum elimination rate and Mi constant The formula:

[0083]

[0084] The baseline metabolic parameters selected by the system based on the genotyping data are reflected in the above-mentioned... , and Individualized specific assignment; for ease of explanation, a specific simulation calculation example is given below: assume that patient A takes the medicine at 08:00, the initial amount of drug in the absorption chamber is recorded as 100 units, and the initial amount of parent drug in the systemic circulation is 0;

[0085] At 09:00, the absorption chamber volume decreased to 60, the systemic circulating mother drug volume increased to 28, and the metabolite volume was 6; at 10:00, the mother drug volume was 24, and the metabolite volume was 10; at 14:00, the mother drug volume was 15, and the metabolite volume was 12; then the instrument response coefficient was mapped to a reference peak area sequence.

[0086] In terms of individualized settings, the system selects baseline metabolic parameters based on genotype. For example, standard enzyme activity parameters can be set for common metabolizers; for known slow metabolizers, the elimination rate is appropriately reduced in the ideal baseline; and for known fast metabolizers, the baseline metabolic rate is increased. It is important to emphasize that the ideal baseline always corresponds to the patient's normal medication status within their inherent genetic background, rather than all patients sharing the same standard curve. At the same time, when constructing this baseline, it is explicitly stated that sampling interference parameters are not superimposed, that is, the instrument detection process is assumed to be ideal, with complete peak shape and no ion suppression, so that all sampling-side abnormalities can be uniformly handled in the perturbation simulation stage.

[0087] Without this time-progression mechanism, using only a static threshold, such as an abnormality if the parent drug is below a certain value, will frequently result in false alarms in fast-metabolizing patients; conversely, it may result in missed alarms in slow-metabolizing patients. Therefore, this implementation method embeds individual genetic information into the benchmark in advance, pre-constructs a pharmacokinetic benchmark that reflects the patient's inherent physiological state, and then measures the true detection residual based on this benchmark.

[0088] As a backup fault tolerance mechanism, when genotyping is missing, the system can first select the median metabolic parameters of the corresponding drug in the population to construct a temporary benchmark and provide a confidence assessment label for the missing gene information.

[0089] When there is an uncertain interval in the dosing time stamp, such as when the patient can only recall taking the medication between 7:00 and 9:00 in the morning, the system can generate multiple candidate baselines within this interval and select the one that best matches the sampling time for subsequent analysis, while retaining the manual review prompt; when the prescription is adjusted and the adjustment spans multiple sampling cycles, the system should proceed separately for different time periods to avoid calculation aliasing of the old dose and the new dose in the same time-series evolution baseline;

[0090] Patient A subsequently underwent genetic testing, which revealed a rapid metabolism-related genotype. At the next follow-up visit, the system no longer used the baseline of the general metabolic population but automatically increased the metabolic rate in the ideal baseline, ensuring that low concentrations but normal proportions were not misinterpreted as missed doses. The purpose of this step was to make the baseline model as close as possible to the patient's inherent physiological state, thereby achieving interpretability and stability in subsequent residual analysis.

[0091] In a preferred embodiment of the present invention, the theoretical abnormal simulation state includes a compliance abnormal simulation state, a metabolic abnormal simulation state, and an interference abnormal simulation state. The parameter injection module includes: a violation injection unit, used to introduce the preset compliance abnormal parameters into the ideal pharmacokinetic benchmark, wherein the preset compliance abnormal parameters include missed dose parameters and / or incorrect dose parameters, so as to generate the compliance abnormal simulation state.

[0092] A metabolic injection unit is used to introduce the preset metabolic abnormality parameters into the ideal pharmacokinetic baseline, wherein the preset metabolic abnormality parameters include enzyme induction parameters and / or enzyme inhibition parameters, to generate the metabolic abnormality simulation state; an interference injection unit is used to introduce the preset acquisition interference parameters into the ideal pharmacokinetic baseline, wherein the preset acquisition interference parameters include ion inhibition parameters, to generate the interference abnormality simulation state.

[0093] This embodiment provides a parameterized anomaly injection mechanism. Specifically, after the aforementioned individualized ideal benchmark has been established, if the real data is directly compared with the benchmark, it can only show how much it has deviated, but it cannot know what kind of anomaly it deviates from. Therefore, this embodiment pre-mathematicizes the common causes of anomalies in outpatient clinics and actively injects them into the ideal benchmark to form a comparable theoretical anomaly template library.

[0094] Specifically, the violation injection unit is used to simulate compliance abnormalities; the missed dose parameter can be understood as the failure to effectively administer medication at a predetermined dosing time. At the algorithm simulation execution level, the system forces the input absorbed dose variable within the time step corresponding to the predetermined dosing time to be set to zero in order to complete the injection of the missed dose parameter. As a result, the concentration of the parent drug drops rapidly, while the metabolites show relatively mild changes due to the residue of the preceding drug and the formation lag; the wrong dosing parameter can simulate early dosing, delayed dosing, or repeated dosing.

[0095] For example, ideally, if the medication is taken at 08:00, the peak area of ​​the parent drug at 10:00 should be 90. If the parameter for a missed dose is injected, the parent drug template at 10:00 can drop to 50 and the metabolite to 28. If the parameter for delayed medication is injected two hours later, the parent drug template at 10:00 may be even lower, while the peak at 14:00 may show a delayed peak. In this way, the system can distinguish between two different types of compliance characteristics: overall low levels and time misalignment.

[0096] The metabolic infusion unit is used to simulate physiological or combined drug-induced changes in metabolic capacity; enzyme induction parameters accelerate the conversion and clearance of parent drug into metabolites, resulting in a faster decline in the parent drug peak and an increase in the proportion of metabolites; enzyme inhibition parameters, on the contrary, increase the effective response during the retention time of the parent drug and result in relatively insufficient metabolite generation.

[0097] For example, if enzyme induction parameters are injected into patient A, the parent drug template at 10:00 drops from 90 to 58, while the metabolite increases from 33 to 46; if enzyme inhibition parameters are injected, the parent drug at 10:00 can rise to 108, while the metabolite decreases to 25. Thus, the template library no longer only shows abnormalities in high and low values, but also exhibits the characteristic of a reversed parent drug-metabolite ratio.

[0098] Interference injection units are used to simulate non-physiological abnormalities in the sampling and detection chain; a typical example is ion suppression, which is matrix co-elution occurring near the retention time window, resulting in a localized suppression of the target ion response; in this case, a decrease in peak area does not necessarily represent a decrease in the actual drug concentration.

[0099] During simulation, the peak area within a certain time window can be proportionally compressed, and slight retention time drift or peak distortion can be superimposed.

[0100] Specifically, the interference injection unit simulates the ion response reduction caused by matrix co-elution by multiplying the peak area integral result of the corresponding retention time window by a random decay coefficient between 0.5 and 0.9.

[0101] For example, at 10:00, the parent drug was reduced from 90 to 72, and the metabolite was reduced from 33 to 25, but the pharmacokinetic time trends of both did not show a large, continuous, monotonous decrease as with missed doses; thus, the collected interference templates have the characteristics of local distortion rather than systemic metabolic changes.

[0102] If the three types of parameters are not injected separately, the system may project all abnormalities onto a single low concentration direction; in extreme cases, ion-inhibited samples may show low values ​​just like missed dose samples, leading to false alerts to patients, while the actual problem lies in the sample quality.

[0103] As a backup fault tolerance mechanism, when the same patient may have two or more abnormal factors at the same time, the system can allow template superposition to generate a composite simulation state, such as mild missed dose + mild ion inhibition.

[0104] When the injected template exceeds the reasonable physiological range, such as when a negative concentration occurs, the metabolite far exceeds the physical upper limit, or there is a reverse peak that does not conform to the dosing sequence, the system will automatically truncate or discard the template to avoid contaminating the template library. When a drug has a complex metabolic pathway and there is more than one main metabolite, sub-templates can be created for each metabolite and then fused according to drug specific weights.

[0105] Patient A's follow-up sample showed a significant decrease in parent drug levels but an increase in metabolites. The system compared the actual residual with templates for missed doses, incorrect doses, enzyme induction, and ion inhibition. It found that the residual was closest to the enzyme induction template, so it prioritized prompting the clinical examination to indicate whether a new combination drug for inducing enzymes had been added. The purpose of this mechanism is to transform common abnormal causes in clinical experience into calculable and comparable templates, thereby achieving fine-grained decomposition of abnormal sources.

[0106] In a preferred embodiment of the present invention, the difference extraction module is used to first perform time-retention alignment, peak area normalization, and filtering processing to suppress baseline drift and high-frequency noise on the multidimensional time-series data in sequence, and then generate the actual residual vector.

[0107] This embodiment provides a preprocessing and extraction mechanism for real residual vectors. Specifically, after the template injection is completed, if the original detection data is directly subtracted from the ideal benchmark, slight drift in retention time, changes in instrument gain, and high-frequency noise will be mistakenly identified as pharmacokinetic abnormalities. Therefore, this embodiment adds a normalized preprocessing chain before generating real residuals.

[0108] In detail, the first step is to perform retention time alignment; the system corrects the actual peak position of the current batch of samples based on the reference windows of the target parent drug peak and metabolite peak; a simplified deduction can be used: the ideal window is 5.2 min, while the actual peak position falls at 5.35 min;

[0109] The system first determines whether the drift amount of 0.15 min is not greater than the preset allowable drift threshold. If it is not greater than the allowable drift threshold, the peak is shifted back to the reference window and then integrated. If it is greater than the allowable drift threshold, it is marked as alignment failure. This step can avoid false anomalies caused by a slight shift in the retention time window, which could lead to a sudden drop in the peak area integral value.

[0110] The second step is to perform peak area normalization. Since different batches of samples may have different injection volumes, ion source states, or instrument sensitivities, the system needs to map the peak areas to a uniform scale.

[0111] Specifically, scaling is performed using the internal standard peak area or the correction factor of the same batch of standard samples; for example, if the original peak area of ​​the parent drug in patient A's test is 8200, while the standard response of the same batch is 10% higher than theoretical, then it is recorded as 7455 after normalization.

[0112] Metabolites are processed in the same way; if a simpler relative proportion method is used, the peak areas of the parent drug and metabolites at each time point can be divided together by the total response at that time point to obtain a normalized sequence that is less sensitive to instrument fluctuations.

[0113] The third step is to perform filtering to suppress baseline drift and high-frequency noise; this can be understood as smoothing the original time series, preserving the slowly changing pharmacokinetic trend while weakening local spikes.

[0114] In practice, the system first uses a one-dimensional moving median filter algorithm to traverse the time series to remove randomly occurring isolated high-frequency spike noise. It then estimates the low-frequency baseline of the time series based on least squares polynomial fitting and subtracts it, thereby purifying the true response integral trajectory.

[0115] For example, in patient A's 10:00 sample, a transient high-frequency noise interference occurred near the target retention time, causing the peak area of ​​the parent drug to increase by 10 units for a short time; after filtering, the high-frequency noise was effectively suppressed, while the true peak envelope was still preserved; the differential extraction module then subtracted the preprocessed true sequence from the ideal benchmark point by point to generate a more stable real residual vector.

[0116] Without this preprocessing chain, although the template library established in the previous implementation is interpretable, the large number of non-target perturbations in real data will cause the template similarity to be dominated by noise; especially when laboratory equipment runs across days, slight retention time drift is very common, and without alignment, normal samples are easily misjudged as abnormal samples.

[0117] As a backup fault-tolerance mechanism, when retention time alignment fails but the mass-to-charge ratio and peak shape can still confirm the target object, the system can use wide-window reintegration and reduce the confidence score.

[0118] When the normalization denominator is too small, for example, when the internal standard peak is not detected or the total response is close to zero, the sample will not participate in the automatic classification and will directly flow into the resampling process; when the peak area loss after filtering exceeds the preset upper limit, it indicates that the smoothing is too strong, the system will restore the result of the previous step and reprocess it with weaker filtering parameters.

[0119] Patient A's chromatograms from two follow-up visits on Monday and Friday had retention times that were 0.08 min and 0.12 min later, respectively, due to differences in instrument status. After alignment and normalization, the true difference between the two samples was mainly due to the time sequence changes of the parent drug / metabolite, rather than differences in instrument status. The purpose of this step is to first clean up the correctable technical noise in the detection chain, thereby achieving higher sensitivity of the real residual vector to clinical abnormalities.

[0120] In a preferred embodiment of the present invention, the coupling decision module is used to calculate the dynamic time warp distance and Mahalanobis distance between the actual residual vector and each of the theoretical residual vectors in the theoretical residual matrix, and to convert the dynamic time warp distance and the Mahalanobis distance into classification similarity by calculating the reciprocal of the distance value or applying a negative exponential function, thereby generating classification similarity values ​​corresponding to the compliance abnormality state, the metabolic abnormality state and the acquisition interference state, respectively.

[0121] This embodiment provides a similarity calculation mechanism based on dual distance coupling. Specifically, after the aforementioned real residuals have been preprocessed, if only a single Euclidean distance is used for matching, it is often prone to failure in two complex situations: First, when there is a slight misalignment or shift in the time axis, the time series curves with highly similar morphological features will be incorrectly amplified in terms of distance measurement.

[0122] Second, when there is objectively physiologically related covariance between different dimensional features, simply performing a dimension-by-dimensional difference operation is difficult to effectively characterize typical multidimensional structured joint patterns such as the decrease in parent drug concentration accompanied by the increase in metabolite concentration; therefore, this implementation method uses dynamic time-regulated distance and Mahalanobis distance in combination.

[0123] Specifically, dynamic time warping distance is used to measure the temporal morphological similarity of sequences. Even if there is a slight temporal shift or delay at the time of the abnormal event, a better alignment relationship can be found by locally stretching and compressing the time axis. For example, if patient A actually missed a dose the previous night, while the missed dose set in the template occurred in the early morning, the two time positions are not completely consistent, but their residual morphology both show a sudden drop in parent drug and a slow drop in metabolites.

[0124] Dynamic time warping can accurately identify and match sequences with similar morphological changes but temporal shifts. This can be illustrated with simplified numerical values: the distance between the actual residual and the missed dose template is 1.6, with the enzyme-induced template it is 2.8, and with the acquisition interference template it is 3.1, indicating that it is closer to the missed dose in terms of temporal morphology.

[0125] Mahalanobis distance is used to measure the statistical structural similarity of the joint distribution of multiple variables; it takes into account the variance of each dimension and the correlation between dimensions; in engineering implementation, the system extracts the distribution characteristics of various abnormal simulation states in the theoretical residual matrix and calculates the multidimensional covariance matrix corresponding to each abnormal sub-category such as enzyme induction and ion inhibition.

[0126] When calculating the Mahalanobis distance between the actual residual vector and a certain theoretical residual vector, the system subtracts the actual residual vector from the theoretical residual vector by equal dimensions to obtain the deviation vector in column vector form, and extracts the inverse matrix of the covariance matrix of the corresponding category; the system multiplies the transpose row vector of the deviation vector on the left by the inverse matrix, and then multiplies it on the right by the deviation vector itself in column vector form, and obtains the square root of the square value of the distance scalar through this dimension matching structure.

[0127] Through this structured matrix flow rule, the linkage between multiple variables can be accurately captured; for example, in enzyme-induced templates, the decrease of parent drug is often accompanied by the increase of metabolites, and this negative correlation structure will be reflected in the covariance; if the actual residuals also have similar linkage, then even if the absolute deviation of a certain dimension does not exceed the preset tolerance range, the Mahalanobis distance can still be relatively small.

[0128] Using simplified numerical representation, the distance between the actual residual and the missed dose template is 2.4, the distance between the actual residual and the missed dose template is 1.2, and the distance between the actual residual and the missed dose template is 2.9, indicating a better match in multidimensional statistical relationships.

[0129] The system then maps the two distances to classification similarity; the mapping method can be either taking the reciprocal or applying a negative exponential function.

[0130] For ease of understanding, let's assume we use... In the above example, the similarity of missed doses obtained by dynamic time warping is approximately 0.20, and the similarity of enzyme induction is approximately 0.06; the similarity of missed doses obtained by Mahalanobis distance is approximately 0.09, and the similarity of enzyme induction is approximately 0.30.

[0131] The system can fuse data according to preset weights, such as a morphological similarity weight of 0.4 and a statistical structure similarity weight of 0.6, resulting in a final missed service classification similarity of [missing service classification similarity]. Enzyme-induced classification similarity is Therefore, this sample is closer to the category of metabolic abnormality.

[0132] If only dynamic time warping is used, local peak misalignment caused by ion suppression may sometimes be misinterpreted as compliance abnormality due to time misalignment; if only Mahalanobis distance is used, the actual missed dose cases with early and late shifts on the time axis will be further away; therefore, this implementation method uses two types of distances to complement each other, so that the classification considers both the temporal morphology and the joint distribution relationship between the parent drug and metabolites.

[0133] As a backup fault-tolerance mechanism, when the number of template samples of a certain type is insufficient, resulting in instability of the covariance matrix, the system can diagonally enhance or backtrack the covariance in the Mahalanobis distance to a weighted Euclidean distance; when the number of continuous sampling points does not reach the preset lower limit and is insufficient to give full play to the advantages of dynamic time warping, its weight can be reduced and the statistical distance weight can be increased.

[0134] When there is a significant conflict in the similarity directions after two distance mappings, such as one strongly pointing to missed service and the other strongly pointing to acquisition interference, the system retains the conflict feature for subsequent mixed anomaly judgment.

[0135] Patient A had an early follow-up visit after a nocturnal attack. Because the blood draw was delayed by about two hours compared to usual, the actual residual was close to the missed dose template in terms of time, but the ratio of parent drug to metabolite was more consistent with enzyme induction. After dual-distance coupling, the system will not determine the missed dose based solely on the time misalignment, but will provide a comprehensive classification similarity value.

[0136] The purpose of this step is to improve the accuracy of multidimensional data judgment between different anomaly categories, thereby achieving a more reliable basis for technical classification.

[0137] In a preferred embodiment of the present invention, the decision rule of the coupled decision module is configured as follows: determining whether each of the generated classification similarity values ​​is less than a preset decision threshold, wherein the decision threshold and the preset proximity threshold are pre-configured and calibrated based on the consistency rate between automatic classification and manual decision conclusions in historical review samples; if each of the classification similarity values ​​is less than the decision threshold, then the state evaluation result is determined and output as the state to be manually reviewed.

[0138] If there is a classification similarity value that is not less than the decision threshold, then the difference between the extracted maximum classification similarity and the second largest classification similarity is calculated; when the difference is not greater than a preset proximity threshold or there are ties in the maximum classification similarity, the state evaluation result is determined and output as the mixed abnormal state.

[0139] When the difference is greater than the proximity threshold and the maximum classification similarity is unique, the state subdivision judgment is performed based on the abnormal state category corresponding to the maximum classification similarity as the initial abnormal category: when the initial abnormal category is the compliance abnormal state, the corresponding missing or incorrect dosage sub-state of the compliance abnormal state is output; when the initial abnormal category is the metabolic abnormal state, the corresponding metabolic abnormal state is output; when the initial abnormal category is the acquisition interference state, the corresponding interference verification sub-state of the acquisition interference state is output.

[0140] This embodiment provides a multi-level judgment and anomaly fallback mechanism. Specifically, after the aforementioned similarity has been calculated, directly taking the maximum similarity as the final conclusion, although simple to implement, is prone to two types of problems in actual outpatient work: one is that the similarity with all preset templates is lower than the set threshold, and forced classification will lead to misjudgment; the other is that the similarity with at least two preset templates meets the threshold and the difference is small, and directly performing a single exclusive classification will mask compound anomalies. Therefore, this embodiment breaks down the judgment into three levels: whether it can be automatically judged, whether it is a mixed state, and whether it can be further subdivided.

[0141] Specifically, the system first checks whether all classification similarities are below the decision threshold. If the decision threshold is 0.60, and the similarity of compliance abnormality is 0.41, the similarity of metabolic abnormality is 0.48, and the similarity of collection interference is 0.37, then all three are below the threshold. The system does not perform forced classification and directly outputs the status pending manual review.

[0142] This means that the sample is neither a typical missed dose nor a typical metabolic abnormality or detection interference. There may be special circumstances outside the template library, such as the patient recently experiencing acute changes in liver function, sample preprocessing errors, or errors in prescription record entry.

[0143] If at least one similarity is not lower than the decision threshold, the system then compares the maximum value with the second largest value. Let compliance abnormality be 0.78, metabolic abnormality be 0.74, and acquisition interference be 0.30. If the proximity threshold is set to 0.05, the difference between the maximum value and the second largest value is 0.04, which is not greater than the proximity threshold. Therefore, the mixed abnormality state is output.

[0144] The results indicate that the sample is likely to carry two types of abnormal features simultaneously, such as the patient having both mild missed doses and metabolic induction factors. If the compliance abnormality is 0.81, the metabolic abnormality is 0.63, the acquisition interference is 0.28, the difference is 0.18, and the maximum value is unique, then the system enters state subdivision.

[0145] When the state is subdivided, if the initial anomaly category is compliance anomaly, the system further compares the missed service sub-template with the wrong service sub-template; for example, if the similarity between the missed service sub-template and the wrong service sub-template is 0.83 and the similarity between the wrong service sub-template and the wrong service sub-template is 0.66, then the missed service sub-state is output; if the similarity between the wrong service sub-template and the wrong service sub-state is higher, then the wrong service sub-state is output.

[0146] If the initial abnormality category is metabolic abnormality, the metabolic abnormality status can be directly output for clinical further examination to see whether it is biased towards induction or inhibition; if the initial abnormality category is collection interference, the interference verification sub-status will be output, indicating that the sample will be given priority to enter the testing link for verification rather than clinical medication adjustment.

[0147] Without the aforementioned hierarchical rules, the engineering implementation of the system would encounter two obvious drawbacks: first, all boundary samples would be hard-classified, increasing clinical distrust; second, composite abnormal samples would be simplified into a single label, resulting in the loss of key information. This implementation method balances automation rate and robustness by using a sequence of threshold filtering, proximity judgment, and sub-state subdivision.

[0148] As a backup fault tolerance mechanism, when the maximum similarity and the second largest similarity are completely equal, even if both are significantly higher than the threshold, the system still outputs a mixed abnormal state instead of randomly selecting one.

[0149] When the missed service sub-template and the wrong service sub-template are close again in the sub-state segmentation, the system can return to the previous level of compliance exception category and add a compliance type uncertainty mark; when the number of templates for a certain exception category is insufficient or has not passed the verification after recent update, the system can temporarily prohibit the category from entering the automatic final judgment and only retain it as an auxiliary prompt.

[0150] Patient A's classification similarity scores for this follow-up visit were: compliance 0.77, metabolic abnormality 0.75, and data collection interference 0.22. Because the difference between the first two scores was very small, the system output a mixed abnormality status and sent it for manual review, rather than simply identifying the patient as having missed a dose. Similarly, another patient, B, had similarity scores of: compliance 0.84, metabolic abnormality 0.40, and data collection interference 0.18, with a higher rate of missed doses within the compliance sub-status. The system directly provided the missed dose sub-status. The purpose of this mechanism is to set clear boundary conditions for automatic judgment, thereby achieving the separate processing of typical samples, boundary samples, and composite samples.

[0151] In a preferred embodiment of the present invention, the management feedback module is specifically configured to: generate the medication reminder instruction in response to the status assessment result being the missed or incorrect dosage sub-state; generate the dosage adjustment suggestion instruction or the medication change reminder instruction for medical staff to review in response to the status assessment result being the metabolic abnormal state; generate the resampling instruction in response to the status assessment result being the interference review sub-state; and generate the manual review instruction in response to the status assessment result being the pending manual review state or the mixed abnormal state.

[0152] This embodiment provides a feedback linkage mechanism from status to management action; specifically, after the aforementioned judgment result has been formed, if the system only outputs an abnormal label without converting it into a specific executable action, the outpatient, pharmacy, and laboratory departments still need to manually interpret it a second time, making it difficult to form a closed loop; therefore, this embodiment directly maps different abnormal states to different management instructions.

[0153] Specifically, when the status result is a missed dose or wrong dose, the system generates a medication reminder instruction; this instruction can be sent to the patient follow-up terminal, SMS platform or nurse station follow-up list. The content does not directly give a medical conclusion, but prompts the patient to check the recent medication time, missed doses and medication records.

[0154] For example, after patient A is identified as missing a dose, the system automatically generates a reminder to check the medication taken last night and this morning and to continue recording it as prescribed by the doctor. At the same time, the two most recent abnormal records are displayed on the medical staff's terminal, which is convenient for follow-up nurses to confirm by phone.

[0155] When the status result is a metabolic abnormality, the system generates a dosage adjustment suggestion or medication change prompt for medical staff to review.

[0156] The emphasis here is on providing a platform for healthcare professionals to review the data, rather than the system directly replacing doctors in making the final decision. For example, if patient A's samples repeatedly show a high degree of similarity to the enzyme-induced template, the system can mark the sample on the doctor's workstation and suggest a review to determine if there is metabolic enhancement, assess the possibility of increasing the dosage, or adjust the combination therapy regimen. If the current medication is nearing its safety limit and still cannot maintain the therapeutic window, a medication switching prompt can be given simultaneously.

[0157] When the status result is the interference verification sub-status, the system generates a resampling instruction. This instruction will return to the laboratory task queue, indicating the problem sample number, the suspected ion inhibition time window, and the recommended resampling time. For example, if the system identifies that a certain sample from patient A is more likely to be ion inhibition than a true concentration decrease, it will not trigger a missed dose reminder on the patient's end, but will first arrange for a new blood sample or a new analysis.

[0158] When the status result is pending manual review or a mixed abnormal status, the system generates a manual review instruction. This instruction will summarize the actual residual, the two closest theoretical templates, the most recent prescription changes, and sampling link information for doctors, clinical pharmacists, or laboratory technicians to review together. This can prevent boundary samples from being mishandled by automated processes without supervision.

[0159] This feedback and linkage mechanism transforms the status assessment results into executable business instructions, thereby achieving a closed-loop process from data analysis to business management;

[0160] As a backup fault tolerance mechanism, when the patient's contact information becomes invalid, resulting in the inability to deliver the medication reminder, the system can forward the reminder to the nurse station's pending tasks; when the dosage adjustment suggestion conflicts with the current prescription contraindication rules, such as when the patient's liver function is abnormal or has reached the recommended upper limit, the system will only retain the medication change prompt and will no longer provide incremental suggestions;

[0161] If the resampling task is not completed within the preset time limit, the system will remind the laboratory department again and notify the outpatient department at the same time to avoid the old samples being used for subsequent decision-making; when the manual review queue is backlogged, the system can sort the priority according to the frequency of recent epileptic seizures, drug risk level, etc.

[0162] When Patient A is first identified as missing a dose during the three-month follow-up, the system sends a medication check reminder to the patient's end; when the second time the patient is identified as having a metabolic abnormality, the system changes to sending a dose review suggestion to the doctor's workstation; if a third time a mixed abnormality occurs, the system summarizes the previous three records to generate a manual review task; the purpose of this mechanism is to allow the classification results to directly drive subsequent business actions, thereby achieving integrated linkage of outpatient, pharmacy and laboratory processes.

[0163] In a preferred embodiment of the present invention, the management feedback module is further configured to receive external review input results for manual review instructions, and update the individual characteristic metabolic parameters in the ideal pharmacokinetic benchmark and synchronously and adaptively correct the decision threshold and the proximity threshold based on the external review input results, so as to provide a parameter determination basis for iterative optimization for subsequent system state assessment.

[0164] This embodiment provides a mechanism for manual review feedback and parameter adaptive update. Specifically, in the aforementioned closed loop, if the system lacks a response step for collecting and integrating external review results after each output of a manual review task, the model parameters will remain stagnant in the initial state for a long time, making it difficult to dynamically adapt to changes in patient medication metabolism and deviations in real data from multiple departments.

[0165] Therefore, this implementation method provides structured feedback of the manual review conclusions to the system and writes them back to the system, which are then used to correct individual baselines and decision-making boundary parameters in multiple dimensions.

[0166] Specifically, manual review results can come from structured input by doctors, clinical pharmacists, or laboratory technicians in the workstation; for example, for a mixed abnormal task of patient A, the doctor finally confirmed that he had recently added an enzyme-inducing combination drug and that the patient's medication record was basically standardized; the laboratory technician also confirmed that there was no obvious ion inhibition in this sample.

[0167] After receiving the result, the system first updates the individual characteristic metabolic parameters in the patient's ideal pharmacokinetic baseline, so that the baseline no longer uses the original ordinary metabolic rate, but changes to a metabolic level that is closer to the current combination condition.

[0168] In the simplified extrapolation, if the ideal peak area of ​​the parent drug at 10:00 is 90 on the original baseline, and the patient is found to be in a high metabolic state for a long time after review, the subsequent baseline can be adjusted to 78; in this way, the same actual detection value of 82 will no longer be a low value deviation exceeding the preset residual threshold in the next evaluation.

[0169] In addition to individual baselines, the system can also synchronously and adaptively adjust the decision threshold and proximity threshold. For example, if in the past period, patients with abnormal compliance and metabolic abnormalities in our hospital have frequently shown similarity and were eventually identified as metabolic abnormalities by manual intervention, it indicates that the existing proximity threshold may be set higher than a reasonable range, causing samples that exceed the preset proportion threshold to be classified as mixed abnormalities.

[0170] The system can fine-tune the proximity threshold from 0.05 to 0.03; conversely, if a large number of samples are automatically judged as a single anomaly, but manual review frequently points out the existence of compound factors, the proximity threshold can be appropriately increased; the judgment threshold can also be gradually modified according to the number of true positives and false positives confirmed by manual review, for example, the automatic final judgment threshold for a certain category can be increased from 0.60 to 0.65 to reduce false judgments.

[0171] In engineering implementation, to prevent excessive fluctuations in the system caused by a single human error input, parameter updates generally adopt a slow correction strategy configuration mechanism; the metabolic parameters of newly harvested individuals do not perform a single full overwrite of historical parameters, but are smoothly integrated according to a set ratio;

[0172] Threshold updates also employ sliding window statistics; for example, the consistency rate between the automatic judgment and the manual conclusion is calculated by continuously collecting the most recent 50 reviewed samples, and then a decision is made on whether to adjust the threshold; this aims to avoid the oscillation of the system's global configuration rules caused by short-lived and accidental individual events.

[0173] Without the aforementioned adaptive feedback mechanism for manual review results, the system can only rely on the static model to perform basic judgments and cannot achieve long-term parameter calibration and self-advanced iterative optimization of discrimination accuracy based on multiple rounds of feedback.

[0174] For long-term management of antiepileptic systems, patient metabolic status, combined medication and sample quality characteristics exhibit significant periodic and sporadic time-varying features, and static configuration parameters will inevitably weaken reliability performance due to changes in the scenario.

[0175] As a backup fault tolerance mechanism, when conflicts arise between manual review results, such as when a doctor marks it as a metabolic abnormality while a laboratory technician marks it as sampling interference, the system does not immediately update the parameters, but instead places the task in a high-priority arbitration list; when the number of review samples available for threshold correction is insufficient, the system maintains the original threshold unchanged and only records the trend.

[0176] When an individual's parameters are modified multiple times in a manner exceeding the preset range, the system can trigger a parameter drift warning, indicating whether there are changes in underlying diseases, alterations in liver and kidney function, or abnormalities in prescription records. When a patient is lost to follow-up for an extended period, resulting in an excessively long update interval, the system can re-establish a new baseline according to the initial procedure when follow-up resumes.

[0177] Patient A was diagnosed with mixed abnormality twice during the first three follow-up visits. After manual review, it was confirmed that the main reason was the enhanced metabolism caused by the addition of new combination drugs. Based on this, the system lowered the expected level of the parent drug in the individual baseline and appropriately tightened the boundary for the judgment of mixed abnormality.

[0178] During the fourth follow-up visit, similar data can be directly identified as metabolic abnormalities, without repeatedly entering the manual review and intervention process. The purpose of this mechanism is to continuously correct the evolution model using real external multi-feedback data, thereby achieving closed-loop optimization of the system's full-link performance in long-term diagnosis and treatment scenarios.

[0179] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A medication evaluation and management system for antiepileptic patients based on chromatography-chromatographic combined data, characterized in that, include: The data acquisition module is used to collect multidimensional time-series data generated by chromatography-mass spectrometry detection of the parent antiepileptic drug and its metabolites in the patient's biological samples, and to obtain the patient's prescription dosage, administration timestamp, weight and genotyping data from the electronic medical record system; The baseline reconstruction module is used to construct an ideal pharmacokinetic baseline based on the prescribed dosage, administration time stamp, body weight, and genotyping data, using a compartmental model and the Michaelis-Menten kinetic rule, and to generate reference concentration time series and ratio curves of the parent drug and metabolites. The parameter injection module is used to perform perturbation simulation on the ideal pharmacokinetic benchmark by using preset compliance abnormal parameters, preset metabolic abnormal parameters and preset acquisition interference parameters obtained from the system preset abnormal template library, respectively, so as to generate the corresponding theoretical abnormal simulation state. The difference extraction module is used to generate a real residual vector based on the difference between the multidimensional time series data and the ideal pharmacokinetic benchmark, and to subtract each of the theoretical abnormal simulation states from the ideal pharmacokinetic benchmark to obtain multiple theoretical residual vectors, and to combine the multiple theoretical residual vectors to form a theoretical residual matrix. The coupling decision module is used to calculate the similarity between the actual residual vector and each of the theoretical residual vectors in the theoretical residual matrix, and output the state evaluation result based on the calculation result. The state evaluation result includes at least one of the following: compliance abnormal state, metabolic abnormal state, acquisition interference state, mixed abnormal state, and state awaiting manual review. The management feedback module is used to generate medication reminder instructions, dosage adjustment suggestion instructions, medication change reminder instructions, resampling instructions, or manual review instructions based on the status assessment results, and send the instructions to the corresponding medical workstation or patient interaction terminal for feedback execution.

2. The antiepileptic patient medication evaluation and management system based on chromatography-chromatographic coupling data according to claim 1, characterized in that, The multidimensional time-series data includes parent drug data and metabolite data; both the parent drug data and the metabolite data include time-series data formed by the changes in retention time, mass-to-charge ratio, and peak area integral with continuous sampling time.

3. The antiepileptic patient medication evaluation and management system based on chromatography-chromatographic coupling data according to claim 1, characterized in that, The baseline reconstruction module is used to solve the equations describing pharmacokinetics over time to generate the ideal pharmacokinetics baseline. The ideal pharmacokinetics baseline is constructed by using baseline metabolic parameters that match the genotyping data and by setting it to an ideal physical detection state without introducing the preset acquisition interference parameters.

4. The antiepileptic patient medication evaluation and management system based on chromatography-chromatographic coupling data according to claim 1, characterized in that, The theoretical anomaly simulation states include compliance anomaly simulation states, metabolic anomaly simulation states, and disturbance anomaly simulation states. The parameter injection module includes: The irregular injection unit is used to introduce the preset compliance anomaly parameters into the ideal pharmacokinetic benchmark, wherein the preset compliance anomaly parameters include missed dose parameters and / or incorrect dose parameters, so as to generate the compliance anomaly simulation state. A metabolic infusion unit is used to introduce the preset metabolic abnormality parameters into the ideal pharmacokinetic baseline, wherein the preset metabolic abnormality parameters include enzyme induction parameters and / or enzyme inhibition parameters to generate the metabolic abnormality simulation state. An interference injection unit is used to introduce the preset acquisition interference parameters into the ideal pharmacokinetic benchmark, wherein the preset acquisition interference parameters include ion suppression parameters to generate the interference abnormality simulation state.

5. The antiepileptic patient medication evaluation and management system based on chromatography-chromatographic coupling data according to claim 1, characterized in that, The differential extraction module is used to first perform time-keeping alignment, peak area normalization, and filtering processing to suppress baseline drift and high-frequency noise on the multidimensional time-series data in sequence, and then generate the actual residual vector.

6. The antiepileptic patient medication evaluation and management system based on chromatography-chromatographic coupling data according to claim 1, characterized in that, The coupling decision module is used to calculate the dynamic time warp distance and Mahalanobis distance between the actual residual vector and each of the theoretical residual vectors in the theoretical residual matrix, respectively. By calculating the reciprocal of the distance value or applying a negative exponential function for mapping processing, the dynamic time warp distance and the Mahalanobis distance are converted into classification similarity, thereby generating classification similarity values ​​corresponding to the compliance abnormal state, the metabolic abnormal state and the acquisition interference state, respectively.

7. The antiepileptic patient medication evaluation and management system based on chromatography-chromatographic coupling data according to claim 6, characterized in that, The decision rule configuration of the coupled decision module is as follows: determine whether the generated classification similarity values ​​are all less than a preset decision threshold. The decision threshold and the preset proximity threshold are pre-configured and calibrated based on the consistency rate between automatic classification and manual decision conclusions in historical review samples. If all the classification similarity values ​​are less than the decision threshold, the state evaluation result is determined and output as the state to be manually reviewed. If there is a classification similarity value that is not less than the decision threshold, then calculate the difference between the extracted maximum classification similarity and the second maximum classification similarity. When the difference is not greater than a preset proximity threshold or there are ties for the maximum classification similarity, the state evaluation result is determined and output as the mixed abnormal state. When the difference is greater than the proximity threshold and the maximum classification similarity is unique, the state subdivision judgment is performed based on the abnormal state category corresponding to the maximum classification similarity as the initial abnormal category: when the initial abnormal category is the compliance abnormal state, the corresponding missing or incorrect dosage sub-state of the compliance abnormal state is output; when the initial abnormal category is the metabolic abnormal state, the corresponding metabolic abnormal state is output; when the initial abnormal category is the acquisition interference state, the corresponding interference verification sub-state of the acquisition interference state is output.

8. The antiepileptic patient medication evaluation and management system based on chromatography-chromatographic coupling data according to claim 7, characterized in that, The management feedback module is specifically used for: In response to the status assessment result being the missed or incorrect dose sub-status, the medication reminder instruction is generated; In response to the status assessment result being the metabolic abnormality, a dose adjustment suggestion instruction or a medication change prompt instruction is generated for medical staff to review; In response to the state assessment result being the interference complex sub-state, the resampling instruction is generated; In response to the status assessment result being either the status awaiting manual review or the mixed abnormal status, the manual review instruction is generated.

9. The antiepileptic patient medication evaluation and management system based on chromatography-chromatographic coupling data according to claim 7, characterized in that, The management feedback module is also used to receive external review input results for manual review instructions, and update the individual characteristic metabolic parameters in the ideal pharmacokinetic benchmark based on the external review input results, and synchronously and adaptively correct the decision threshold and the proximity threshold, so as to provide a parameter determination basis for iterative optimization for subsequent system status assessment.

Images