Drug conflict automatic detection and prescription optimization system for senile multi-disease patients

Abstract

The invention relates to the technical field of information processing, in particular to a medicine conflict automatic detection and prescription optimization system for old multi-disease patients. According to the system, a data management module is used for collecting individual data of patients and group data of co-diseased reference groups and receiving a planned medication scheme; the parameter calculation module is used for identifying a target medicine combination with time overlapping in the patient medication record and the planned medication scheme; determining a basic risk index according to the group medication response data, the patient medication record and the work and rest text data; determining a lag risk coefficient according to the physiological indexes and the drug metabolism data; determining group medication associated risk parameters according to atypical reactions and group medication records in the medication reaction data; the risk fusion module is used for determining a drug use conflict risk value based on the basic risk index, the lagging risk coefficient and the group drug use associated risk parameter; and the prescription optimization module is used for generating an optimal medication scheme, so that the generated optimal scheme is more suitable for the actual life of the patient.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, specifically to an automatic detection and prescription optimization system for medication conflicts in elderly patients with multiple comorbidities. Background Technology

[0002] With the accelerating aging of society, the number of elderly patients suffering from multiple comorbidities is increasing, and the use of multiple medications is extremely common. While treating diseases, the combined use of multiple drugs significantly increases the risk of drug interactions. These interactions may lead to reduced efficacy or increased toxicity, causing medication conflicts and seriously threatening patient safety.

[0003] Currently, clinical drug conflict detection primarily focuses on the dosage rationality of a single drug or the direct interaction between two drugs, relying on manual prescription review for assessment. This approach has significant limitations: First, manual review struggles to comprehensively and efficiently handle a patient's complex medication history and dozens of potential drug interactions; second, existing knowledge bases mainly focus on immediate drug incompatibilities, lacking effective means to identify and quantify delayed and hidden medication risks arising from drug metabolic residues, individual patient physiological fluctuations, and atypical reactions within a population; finally, traditional detection methods cannot integrate and utilize massive amounts of historical population medication response data, making it difficult to accurately predict specific medication risks for particular comorbidity groups. Summary of the Invention

[0004] To address the technical problem of lacking effective identification and quantification of delayed and hidden medication risks caused by drug metabolic residues, individual patient physiological fluctuations, and atypical reactions in a population, this invention provides an automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities. The specific technical solution adopted is as follows: This invention proposes an automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities. The system includes: The data management module is used to collect individual patient data and group data of the comorbidity reference group, and to receive the proposed medication plan. Individual data includes medication records, daily routine data and physiological indicators, while group data includes group medication records, medication response data and drug metabolism data. The parameter calculation module, which communicates with the data management module, is used to identify target drug combinations that overlap in time between patient medication records and proposed medication regimens; determine basic risk indicators based on the associated reaction data of target drug combinations in the group medication response data, patient medication records, and daily routine text data; determine the lag risk coefficient based on the temporal fluctuation characteristics of physiological indicators and drug metabolism data; and determine the group medication-related risk parameters based on the frequency of atypical reactions in the medication response data and the distribution characteristics of group medication records. The risk fusion module communicates with the parameter calculation module and is used to determine the medication conflict risk value based on basic risk indicators, lag risk coefficients and group medication association risk parameters. The prescription optimization module communicates with the risk fusion module and is used to call the optimization algorithm to iteratively optimize the proposed medication plan with the goal of minimizing the risk of medication conflict, so as to generate the optimal medication plan.

[0005] Furthermore, identify target drug combinations that overlap in time between the patient's medication records and the proposed medication regimen, including: Obtain the historical medication time series from the patient's medication record and the proposed medication time series from the proposed medication plan to construct a unified medication time axis. The historical medication time axis marks the actual medication time and the drug's duration of action, while the proposed medication time axis marks the planned medication time and the planned duration of action. Obtain the minimum safe dosing interval standard for each drug pair involved in the medication timeline; The effects of different drugs in the medication timeline are compared. If the overlap of the effects of any two drugs exceeds the safety threshold, or the planned medication interval is less than the corresponding minimum safe medication interval, then the two drugs are identified as target drugs and combined to form a target drug combination with overlapping medication times.

[0006] Furthermore, the process of determining the basic risk indicators includes: The statistical data on adverse reactions in the population were analyzed, including the first person to experience an adverse reaction after using the target drug combination, the total number of people in the reference population who experienced adverse reactions after using the target drug alone; The ratio of the first patient to the total number of patients using the medication was calculated as the adverse reaction rate in the combination therapy group; the ratio of the second patient to the total number of patients using the medication was calculated as the adverse reaction rate in the single-drug group. For each group of target drug combinations, a baseline value for group risk is obtained by using a preset group risk calculation function based on the adverse reaction incidence rate of the combination group and the adverse reaction incidence rate of each target drug in the individual drug group. Based on patient medication records and daily routine text data, individual risk correction factors were determined; The product of the individual risk adjustment factor and the group risk benchmark is calculated as the basic risk indicator.

[0007] Furthermore, the process for determining individual risk correction factors includes: Extract the time difference between two consecutive doses of the same drug from historical medication time series to form a time difference dataset; divide the time difference dataset into a predetermined number of continuous intervals, count the number of time differences in each interval and the total number of time differences; calculate the proportion of the number of time differences in each interval to the total number, as the interval time frequency value. The entropy value of all time frequency values ​​in all intervals is calculated using the entropy method. The natural logarithm of the total number of intervals is calculated as the limiting entropy value. The ratio of the entropy value to the limiting entropy value is normalized and used as the medication regularity score. Natural language processing is performed on the daily routine text data. Keywords related to daily routines are extracted through word segmentation. Positive and negative features are identified based on the medical word vector model, and the frequency of feature occurrence is statistically analyzed. The frequency of positive and negative features is input into a pre-trained daily routine scoring model, and a normalized daily routine score is output. The patient's medication adherence score and daily routine score are added together to obtain the total score; the ratio of the total score to the arithmetic mean of the total scores of all patients is calculated as an individual risk correction factor to quantify the differences in individual and group patterns.

[0008] Furthermore, the process of determining the lag risk coefficient includes: Acquire the mean steady-state drug concentration and population metabolic characteristic coefficients from drug metabolism data; and extract the total number of days the target drug was taken from patient medication records; For each target drug, the product of the average steady-state drug concentration and the population metabolic characteristic coefficient is calculated as the baseline metabolic index; the ratio of the baseline metabolic index to the total number of days for the corresponding target drug is calculated as the residual impact coefficient reflecting the potential residual risk of the drug. For any target drug in the target drug combination, the incidence rate of adverse reactions in the single-drug group of the target drug is increased by a preset multiple as a population reference threshold; based on the incidence rate of adverse reactions in the combination group, the population reference threshold, and the residual effect coefficient, the probability of lagged risk is determined. Extract multiple time-series detection values ​​of the same physiological indicator within a preset time period and the preset normal range of the physiological indicator from the physiological indicators; normalize the ratio of the standard deviation of the time-series detection values ​​to the width of the preset normal range of the physiological indicator to obtain the physiological state fluctuation index. The sum of the positive integer 1 and the physiological state fluctuation index is calculated as the adjustment factor; the product of the lagged risk probability and the adjustment factor is calculated as the lagged risk coefficient.

[0009] Furthermore, based on the incidence of adverse reactions in the combined treatment group, the population reference threshold, and the residual effect coefficient, the probability of lagged risk is determined, including: For each target drug combination, based on a preset combination weighting coefficient, the residual influence coefficient of the individual drug in the target drug combination is coupled into a combination residual influence coefficient; Select the maximum population reference threshold for a single drug from the target drug combination; The average difference between the incidence of adverse reactions in the combination group of the target drug combination and the incidence of adverse reactions in the single-drug group of the single target drug in the combination is calculated as the additional risk increment; Based on preset weighting coefficients, the maximum group reference threshold and the additional risk increment are weighted and fused to obtain the superimposed risk value; the product of the superimposed risk value and the combined residual influence coefficient is calculated as the lag risk probability.

[0010] Furthermore, the process for determining the risk parameters associated with group medication use includes: Screen for atypical reactions in the group drug response data, count the number of target patients associated with the target drug combination of atypical reactions and the number of atypical reaction events, and calculate the ratio of the number of atypical reaction events to the number of target patients as the frequency of atypical reaction. The frequency of occurrence of any target drug combination in the medication records of the statistical reference group is calculated; the information entropy of all occurrence frequencies is normalized to obtain the distribution complexity index. The ratio of the frequency of atypical reactions to the distribution complexity index is calculated and used as a parameter for the risk associated with group medication use.

[0011] Furthermore, screening for atypical reactions is achieved through a two-stage exclusion mechanism, including: The system retrieves symptom tags, occurrence times, and associated drug combinations from population medication response data; it also calls a pre-set medication rule base to retrieve standardized tags and corresponding drug combinations for known medication responses; and it calls a pre-set patient comorbidity information base to retrieve the patient's currently active comorbidity types and their typical clinical manifestation tags. First-level exclusion: Match the drug reaction label with the standardized labels of known drug-associated reactions. If a match is found, it is identified as a known reaction and excluded. Second-level exclusion: For the unmatched medication responses remaining after the first-level exclusion, the remaining medication response tags are further matched with the typical clinical manifestation tags of the patient's currently active comorbidities. If a match is found, the response is excluded. The remaining drug reactions after the above two-stage exclusion are called atypical reactions and are stored in the atypical reaction dataset. At the same time, their frequency of occurrence in the population and information on associated drug combinations are recorded.

[0012] Furthermore, the process for determining the risk value of medication conflicts includes: Calculate the arithmetic mean of the basic risk indicators of all target drug combinations in the proposed treatment plan, and use it as the first risk indicator; The product of the first risk indicator and the lagged risk coefficient is calculated to obtain the sum of the individual immediate and lagged risks; the sum is added to the group medication-related risk parameter to obtain the medication conflict risk value.

[0013] Furthermore, with the goal of minimizing the risk of medication conflicts, an optimization algorithm is invoked to iteratively optimize the proposed medication regimen, generating an optimal medication regimen, including: The initial population of individuals is based on the proposed drug administration plan, and the fitness function is the reciprocal of the drug conflict risk value. New populations are generated through selection, crossover, and mutation operations, where mutation operations include random perturbations of drug type, dosage, and administration time. The above process is iteratively executed until the convergence condition is met, and the individual with the highest fitness is finally output as the preferred medication regimen. Throughout the optimization process, all generated medication regimens must meet the pharmacologically safe dosage range and contraindications.

[0014] The present invention has the following beneficial effects: By integrating drug combination-related reactions (population data), medication patterns and lifestyle characteristics (individual data), and temporal fluctuations of physiological indicators (time dimension data) through a parameter calculation module, a multi-dimensional risk assessment system is constructed. Simultaneously, by determining the probability of lagged risks through two levels—basic single-drug risk and incremental risk of combination drugs—the disconnect between population and individual data is addressed, ultimately significantly reducing the assessment error of drug conflict risk values. Furthermore, by constructing a multi-source data fusion analysis platform, a shift from reliance on human experience-based judgment to intelligent automated analysis is achieved, greatly improving detection efficiency and coverage while reducing human oversight. By incorporating patients' individual lifestyle patterns, medication habits, and other adherence characteristics into the risk assessment model, the generated optimized plans are more aligned with patients' actual lives, helping to reduce medication risks caused by missed or incorrect doses. Attached Figure Description

[0015] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a structural example diagram of an automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities, provided as an embodiment of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0018] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0019] The following description, in conjunction with the accompanying drawings, details the specific scheme of the automatic detection and prescription optimization system for medication conflicts in elderly patients with multiple comorbidities provided by this invention.

[0020] Please see Figure 1 The diagram illustrates a flowchart of an automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities, according to an embodiment of the present invention. The method includes: S101: Data management module, used to collect individual patient data and group data of comorbidity reference groups, and to receive proposed medication plans; among which, individual data includes medication records, daily routine data and physiological indicators, and group data includes group medication records, medication response data and drug metabolism data.

[0021] Individual data is personalized information that reflects the medication background and physiological state of a specific patient. The specific patient is determined according to the actual situation, and this embodiment does not make a specific limitation, such as elderly patients aged 60 to 80.

[0022] It should be noted that patient medication records can be obtained from devices such as hospital information systems (HIS), electronic health records (EHR), or smart pillboxes. For example, the patient's medication history details for the past 6 months include: drug name, exact time of each dose, actual dosage, duration of continuous use, and time of discontinuation.

[0023] It should be noted that the daily routine text data can be collected through a combination of structured questionnaires and open-ended responses. The daily routine text data can be divided into regular data (such as sleep start and end times) and irregular data (such as "recently, I often fall asleep at 1 a.m. due to insomnia").

[0024] It should be noted that the specific physiological indicators are determined based on actual needs. For example, some routine biochemical indicators, such as liver and kidney function (ALT, AST, creatinine, blood urea nitrogen), blood glucose, and blood lipids, are sorted by the time of detection to form time-series data. Alternatively, some routine vital signs, such as heart rate, blood pressure, and body temperature, are sorted by the time of detection to form time-series data. In addition, in order to target some special comorbidities, it is also necessary to collect some special indicators, such as the glycated hemoglobin physiological indicator for diabetic patients.

[0025] All physiological indicators are correlated with normal reference ranges (adjusted according to the standards for the elderly population).

[0026] Population data can be collected from publicly available, multi-center clinical databases and drug surveillance networks.

[0027] The reference group is a group that is similar to the patients in terms of gender, age (e.g., no more than 5 years apart), and comorbidity (e.g., both have hypertension and type 2 diabetes).

[0028] Group medication records should include at least the frequency of drug combination use (e.g., the percentage of patients taking nifedipine and metformin simultaneously), the average dosing interval between the two drugs in the combination, the average dose of the same drug in the group, and the average number of consecutive days of use for different drug combinations.

[0029] It should be noted that, assuming there are two drug combinations, even if the two drug combinations use exactly the same types of drugs, if the dosage of any drug is different, the two drug combinations will not be classified as the same drug combination.

[0030] Drug response data records adverse events in the reference group after drug use, including at least a description of the basic symptoms after taking a certain drug (e.g., dizziness), the time of occurrence (a few hours or days after taking the drug), and the severity (mild, moderate, or severe); related drug information: the single drug or combination of drugs that triggered the reaction (excluding reactions clearly caused by overdosing).

[0031] Drug metabolism data should include at least the following metabolic parameters: the average half-life, time to peak concentration, and clearance rate of each drug, as well as metabolic pathway interaction data when drugs are combined (e.g., drug A reduces the clearance rate of drug B by 20%); residual characteristics: the average steady-state drug concentration in the body after long-term use and the residual decay curve after drug withdrawal; and population difference coefficients: the ratio of metabolic parameters between the elderly population and healthy adults (e.g., the half-life of drug A in the elderly population is 1.5 times that in healthy adults).

[0032] The proposed medication plan can be received through an interface or imported from the prescription system, and should include at least the following: drug list: name, specifications, single dose, and number of times to take the drug; medication schedule (e.g., take at 8:00 AM, 8:00 PM, or 30 minutes before meals); medication cycle: start date, expected end date, or course of treatment (e.g., 14 consecutive days); and instructions for combination therapy: the basis for combining different drugs (e.g., combining drug C to enhance the antihypertensive effect of drug A).

[0033] S102: Parameter calculation module, which communicates with the data management module, is used to identify target drug combinations that overlap in time between patient medication records and proposed medication plans; determine basic risk indicators based on the associated reaction data of target drug combinations in the group medication response data, patient medication records, and daily routine text data; determine the lag risk coefficient based on the temporal fluctuation characteristics of physiological indicators and drug metabolism data; and determine the group medication-related risk parameters based on the frequency of atypical reactions in the medication response data and the distribution characteristics of group medication records.

[0034] In this embodiment, the historical medication time series in the patient's medication record and the proposed medication time series in the proposed medication plan are obtained to construct a unified medication time axis. The historical medication time axis marks the actual medication time and the drug's duration of action, while the proposed medication time axis marks the planned medication time and the planned duration of action. The minimum safe dosing interval standard corresponding to each drug pair involved in the medication time axis is obtained. The durations of action of different drugs in the medication time axis are compared. If the duration of action of any two drugs overlaps for more than the safe threshold, or the planned dosing interval is less than the corresponding minimum safe dosing interval, then the two drugs are determined to be target drugs and combined to form a target drug combination with overlapping dosing times.

[0035] Historical medication time series refers to the collection of actual medication information extracted from a patient's past medication records and ordered chronologically, such as "2025-09-01 08:00 took antihypertensive drug A (5mg)" and "2025-09-01 20:00 took hypoglycemic drug B (10mg)". Each record includes the specific drug name, the time of administration, and the actual dosage.

[0036] The proposed medication time series refers to the set of medication plan information to be issued by relevant personnel, including the name of the planned medication, the planned time of administration, and the planned dosage.

[0037] It is understood that the specific method for constructing a unified medication timeline is an existing technical means, which will not be elaborated in this embodiment. For example, the historical medication time series and the proposed medication time series are merged into a continuous timeline in chronological order.

[0038] Drug pairing: refers to any two drugs that may be used simultaneously or sequentially in the medication timeline (such as A in historical medication and C in proposed medication, B in historical medication and B in proposed medication, etc.).

[0039] Minimum safe dosing interval standard: refers to the shortest time interval between two drugs to avoid metabolic conflicts (such as affecting absorption). It should be noted that the specific value of the minimum safe dosing interval standard is determined based on pharmacological studies and clinical guidelines. This embodiment does not make specific limitations. For example, the minimum safe interval between drug A (antihypertensive drug) and drug C (lipid-lowering drug) is 4 hours.

[0040] To accurately determine the duration of overlap in the duration of action, as an example, the duration of action of any two drugs in the medication timeline is examined one by one (including the actual duration of action of the historical drugs and the planned duration of action of the proposed drugs), and the overlap duration is calculated. For example, if the actual duration of action of drug A is "08:00-16:00" and the planned duration of action of drug C is "09:00-23:00", the overlap duration is 7 hours.

[0041] It should be noted that the specific value of the safety threshold is determined according to the actual situation, and this embodiment does not impose a specific limitation. For example, the safety threshold can be 2 hours.

[0042] It is important to understand that when patients use a target drug combination, the core risk level varies due to both individual differences and group medication patterns. On the one hand, group medication response data can reflect the inherent risk of the drug combination and provide an objective risk benchmark. On the other hand, patients' own medication records (such as medication regularity) and lifestyle data (such as lifestyle stability) can amplify or reduce the risk, which needs to be adjusted through individual correction factors. Therefore, it is necessary to further analyze the gap between the general risk of the group and the individualized risk of the patient.

[0043] To accurately determine baseline risk indicators from both general population risk and individual patient risk perspectives, as an example, the following methods are used: First, the number of patients experiencing adverse reactions after using the target drug combination in the population medication response data; the total number of patients in the reference population who experienced adverse reactions after using the target drug alone; second, the number of patients experiencing adverse reactions after using the target drug alone. The ratio of the first patient to the total number of patients is calculated as the adverse reaction rate in the combination group; the ratio of the second patient to the total number of patients is calculated as the adverse reaction rate in the single-drug group. For each target drug combination, based on the adverse reaction rate in the combination group and the adverse reaction rate in the single-drug group of each target drug in the combination, a baseline population risk value is obtained through a pre-defined population risk calculation function. Based on patient medication records and daily routine data, an individual risk correction factor is determined. The product of the individual risk correction factor and the baseline population risk value is calculated as the baseline risk indicator.

[0044] Total number of patients in the reference group: refers to the total number of patients using the target drug combination.

[0045] It should be noted that, in order to ensure the objectivity of the data calculated later, a suitable number of reference subjects (i.e., the reference group) have been selected when selecting the reference group. Therefore, given that the existence of the target drug combination has been confirmed, the total number of people using the drug in the reference group cannot be zero.

[0046] In order to accurately obtain the baseline value of population risk through a pre-defined population risk calculation function, assuming that a target drug combination includes drug A and drug B, the baseline value of population risk can be determined by the following pre-defined population risk calculation function: Group risk benchmark = ; in, This indicates the first person to experience an adverse reaction after using the target drug combination (drug A and drug B) in the group medication response data; This indicates the incidence of adverse reactions in the single-drug group of drug A; This represents the incidence of adverse reactions in the single-drug group of drug B; a and b represent weights, which are set based on the interaction strength of the two drugs in combination. The stronger the interaction, the higher the weight of the incidence in the combination group. For example, a is 0.7 and b is 0.3.

[0047] The population risk benchmark reflects the average risk level of the target drug combination in the reference population.

[0048] To accurately obtain individual risk correction factors, as an example, the time difference between two consecutive doses of the same drug is extracted from historical medication time series to form a time difference dataset. This dataset is divided into a predetermined number of continuous intervals, and the number of time differences within each interval and the total number of time differences are counted. The proportion of time differences in each interval to the total number is calculated as the interval time frequency value. The entropy value of all interval time frequencies is calculated using the entropy method, and the natural logarithm of the total number of intervals is used as the limiting entropy value. The ratio of the entropy value to the limiting entropy value is normalized to obtain the medication regularity score. Natural language processing is performed on the daily routine text data, extracting keywords related to daily routines through word segmentation. Positive and negative features are identified based on a medical word vector model, and the frequency of feature occurrence is counted. The frequencies of positive and negative features are input into a pre-trained daily routine scoring model, outputting a normalized daily routine score. The patient's medication regularity score is added to the daily routine score to obtain a total score. The ratio of the total score to the arithmetic mean of all patients' total scores is calculated as the individual risk correction factor quantifying the differences in individual and group patterns.

[0049] For example, from the patient's historical medication time series, all medication time points for the same drug (such as daily antihypertensive drugs) are selected (sorted by time), the time difference between two adjacent medications is calculated (e.g., "medication taken at 8:00 on September 1st, medication taken at 7:30 on September 2nd, time difference 23.5 hours"), and all time differences are summarized to form a "time difference dataset" (e.g., [23.5h, 24.2h, 22.8h, ...]).

[0050] It should be noted that the specific value of the preset quantity is determined according to actual needs, and this embodiment does not impose a specific limitation. For example, the value range of the preset quantity is usually 6 to 10.

[0051] One common technique is to divide the time into intervals. For example, when the time difference is between 20 and 26 hours, it can be divided into 6 intervals: [20-21h, 21-22h, ..., 25-26h].

[0052] In real-life scenarios, patients cannot take the same medication at exactly the same time, so there will inevitably be time differences. Therefore, the total number of time differences cannot be zero. For new patients with no history of medication, their medication regularity score can be assumed to be the group average, or other initialization strategies can be adopted.

[0053] Entropy method: A mathematical method for measuring the dispersion of data distribution. The entropy method is a well-known technique in the art, and will not be described in detail in this embodiment. The more dispersed the data distribution (the greater the fluctuation of time difference, the more irregular the medication), the higher the entropy value; the more concentrated the distribution (the more stable the time difference, the more regular the medication), the lower the entropy value.

[0054] The limiting entropy is the natural logarithm of the total number of intervals (e.g., for 6 intervals, the limiting entropy = ...). ≈1.79), representing the maximum entropy value when the time difference is completely randomly distributed.

[0055] It is important to understand that a larger entropy value reflects a more dispersed data distribution, more irregular medication administration, and a lower limit of entropy. Conversely, a smaller limit of entropy indicates a smaller time difference in medication administration, a denser distribution of time differences, and a lower degree of uniformity, which also suggests that medication administration is extremely irregular.

[0056] It should be noted that natural language processing and keyword extraction through word segmentation are techniques well known to those skilled in the art, and will not be described in detail in this embodiment.

[0057] The medical word vector model (referring to an AI model trained in advance with a large amount of medical lifestyle text) classifies keywords into positive features (representing regular lifestyles, such as going to bed at 11 pm and waking up at 7 am) and negative features (representing disordered lifestyles, such as insomnia and going to bed at 1 am).

[0058] The pre-trained daily routine scoring model is trained using correlation data between feature frequencies and daily routine stability. The specific model training method is a well-known technique in the art, and will not be described in detail in this embodiment.

[0059] It's important to understand that the higher the frequency of positive features and the lower the frequency of negative features, the closer the daily routine score is to 0, meaning the lower the score, the more regular the daily routine.

[0060] It is important to understand that elderly patients with multiple diseases often have a decreased metabolic capacity, which leads to slow drug clearance from the body. Even after stopping the medication, the drug may still maintain a certain concentration for a period of time, which may interact with subsequent medications. At the same time, fluctuations in physiological indicators (such as blood pressure and blood sugar) can amplify the risk of adverse reactions from this residual drug. Therefore, the effects of drug residues can be combined with the effects of physiological fluctuations to make time-based corrections to the basic risks and also cover the delayed effects after medication.

[0061] In this embodiment, the average steady-state drug concentration and population metabolic characteristic coefficient included in the drug metabolism data are obtained; and the total number of days of taking the target drug is extracted from the patient's medication records; for each target drug, the product of the average steady-state drug concentration and the population metabolic characteristic coefficient is calculated as a baseline metabolic index; the ratio of the baseline metabolic index to the total number of days of taking the corresponding target drug is calculated as a residual impact coefficient reflecting the potential residual risk of the drug; for any target drug in the target drug combination, the incidence rate of adverse reactions in the single drug group of the target drug is increased by a preset multiple as a population reference threshold; based on the incidence rate of adverse reactions in the combination group, the population reference threshold, and the residual impact coefficient, the lag risk probability is determined; multiple time-series detection values ​​of the same physiological index within a preset time period and a preset normal range of the physiological index are extracted from the physiological index; the ratio of the standard deviation of the time-series detection values ​​to the width of the preset normal range of the physiological index is normalized to obtain a physiological state fluctuation index; the sum of the positive integer 1 and the physiological state fluctuation index is calculated as a regulation factor; and the product of the lag risk probability and the regulation factor is calculated as the lag risk coefficient.

[0062] Average steady-state drug concentration: The average drug residue concentration obtained from blood tests after a reference group has been taking the target drug for a long time and has reached a stable period of drug use (the drug concentration in the body no longer fluctuates significantly). It reflects the basic residual level of the drug in the body. The higher the concentration, the greater the risk of residue.

[0063] Population metabolic characteristic coefficient: The ratio of the median half-life of the target drug in the reference population (elderly patients with multiple diseases) to the median half-life of the same target drug in healthy adults (e.g., if the half-life of the elderly population is 1.5 times that of healthy adults, the population metabolic characteristic coefficient = 1.5). A population metabolic characteristic coefficient greater than 1 indicates that elderly patients metabolize drugs more slowly and are more likely to leave drug residues.

[0064] It is important to understand that if the total number of days a target drug is taken is shorter, the drug concentration may not reach a steady state before the drug is suddenly discontinued, which increases the risk of residues and interactions with subsequent medications. The baseline metabolic index comprehensively reflects the inherent drug residue level and the population's metabolic capacity. If the baseline metabolic index is higher, it indicates a higher baseline risk of drug residues in the reference population.

[0065] The residual impact coefficient quantifies the potential impact of drug residues on risk.

[0066] The specific value of the preset multiple is not specifically limited in this embodiment. For example, based on the clinical research consensus that the residual effect of the drug may amplify the risk in pharmacology, the preset multiple can be 2 times.

[0067] To accurately determine the probability of lag risk, as an example, for each target drug combination, based on a preset combination weighting coefficient, the residual impact coefficient of a single drug in the target drug combination is coupled into a combination residual impact coefficient; the maximum population reference threshold for a single drug in the target drug combination is selected; the average difference between the adverse reaction incidence rate of the combination group and the adverse reaction incidence rate of the single target drug group in the combination is calculated as an additional risk increment; based on the preset weighting coefficient, the maximum population reference threshold and the additional risk increment are weighted and fused to obtain the superimposed risk value; the product of the superimposed risk value and the combination residual impact coefficient is calculated as the probability of lag risk.

[0068] The specific value of the preset combination weight coefficient is not specifically limited in this embodiment. For example, it can be set based on the interaction strength of the metabolic pathways of the two target drugs (for example, if drugs A and B compete with liver enzymes, the stronger the interaction, the higher the weight coefficient, usually 0~1). For example, when the interaction strength is medium, the weight coefficient = 0.6.

[0069] It should be noted that the coupling calculation method is a common technique, which will not be elaborated in this embodiment. For example, the combined residual influence coefficient = [drug A residual influence coefficient × drug A weight coefficient + drug B residual influence coefficient × drug B weight coefficient] + [(1 - drug A weight coefficient) × drug A residual influence coefficient × drug B residual influence coefficient].

[0070] The population reference threshold is the risk boundary of a single drug after considering residues. The maximum value of the two is selected as the basic risk benchmark value for the combination. In order to avoid underestimating the basic residue risk, the residue risk of the two drugs used in combination is usually not lower than that of the single drug with higher risk.

[0071] Since the additional risk increment reflects the average risk of the combination drugs compared to the two target drugs used alone, the additional probability of adverse reactions is a risk increment unique to the drug combination. Therefore, the additional risk increment can be expressed by the following formula: Additional risk increment = Adverse reaction rate of the combination drug group - (Adverse reaction rate of drug A monotherapy group + Adverse reaction rate of drug B monotherapy group) ÷ 2.

[0072] It should be noted that the specific value of the preset weight coefficient is not specifically limited in this embodiment. For example, the weight of the maximum group reference threshold is usually higher, such as 0.6, and the weight of the additional risk increment is 0.4.

[0073] It should be noted that the specific value of the preset time period is determined according to actual needs, and this embodiment does not impose a specific limitation. For example, the preset time period is the past 7 days.

[0074] It should be noted that the specific values ​​of the preset normal range of physiological indicators are determined according to actual needs. This embodiment does not make specific limitations. For example, based on industry clinical experience, different physiological indicators have different preset normal ranges. For example, the normal range of systolic blood pressure is 90-140 mmHg.

[0075] It is important to understand that if the standard deviation of the time-series detection values ​​is larger, it reflects the magnitude of the fluctuation and the more severe the fluctuation, then after normalization, the physiological state fluctuation index is obtained. The higher the value, the more severe the physiological fluctuation. The lag risk coefficient quantifies the delayed risk of the combined effect of "drug residue + physiological fluctuation". The higher the lag risk coefficient, the greater the possibility of delayed adverse reactions occurring within a certain period of time after medication.

[0076] It's important to understand that while the adverse reaction risks of a single drug can be obtained through clinical trials or post-marketing surveillance, drug combinations may produce synergistic or antagonistic effects, leading to a significant increase or decrease in risk. Furthermore, individualized risk assessment requires a reference standard; population risk parameters serve as this benchmark. By comparing individual circumstances with the population benchmark, it can be determined whether a patient's risk is higher or lower than the average level.

[0077] In this embodiment, atypical reactions are screened from the group medication response data. The number of target patients associated with the target drug combination and the number of atypical reaction events are counted. The ratio of the number of atypical reaction events to the number of target patients is calculated as the frequency of atypical reaction occurrence. The frequency of occurrence of any target drug combination in the reference group medication records is counted. The information entropy of all occurrence frequencies is normalized to obtain a distribution complexity index. The ratio of the frequency of atypical reaction occurrence to the distribution complexity index is calculated as a group medication association risk parameter.

[0078] Atypical reactions refer to adverse reactions that are inconsistent with known drug safety information, occur very infrequently, or are of high severity.

[0079] The frequency of occurrence reflects the prevalence of atypical reactions in a population.

[0080] It should be noted that the method for calculating information entropy is an existing technical means, which will not be described in detail in this embodiment. The more dispersed the distribution of drug combinations, the higher the entropy value; the more concentrated the drug combinations, the lower the entropy value.

[0081] It should be noted that the distribution complexity index is the normalized information entropy. An entropy of zero means that only one drug combination is used, and the usage frequency of other combinations is all 0. However, in a group medication scenario, it is almost impossible for everyone to use the exact same drug combination, so the distribution complexity index cannot be zero.

[0082] It is important to understand that the higher the frequency of atypical reactions, the more prevalent the atypical reactions are in the population. Furthermore, the lower the distribution complexity index (the lower the entropy value), the more concentrated the drug combination is. This indicates that atypical reactions occur frequently and are associated with multiple drug combinations, thus the higher the risk of drug association in the population.

[0083] To comprehensively screen for atypical reactions, it should be noted that a two-stage exclusion mechanism is used. As an example, the following steps are taken: First, the symptom tags, occurrence times, and associated drug combinations of medication reactions are obtained from the population medication reaction data. Simultaneously, a pre-set medication rule base is accessed to obtain standardized tags and corresponding drug combinations for known medication reactions. Second, a pre-set patient comorbidity information base is accessed to obtain the patient's currently active comorbidity types and their typical clinical manifestation tags. The first stage of exclusion involves matching the medication reaction tags with the standardized tags of known medication-associated reactions. If a match is found, the reaction is considered a known reaction and excluded. The second stage of exclusion further involves matching the remaining unmatched medication reactions with the typical clinical manifestation tags of the patient's currently active comorbidities. If a match is found, the reaction is excluded. The remaining medication reactions after these two stages of exclusion are considered atypical reactions and are stored in the atypical reaction dataset, along with their frequency of occurrence and associated drug combination information within the population.

[0084] Medication reaction symptom labels: such as "dizziness," "rash," etc. Time of occurrence: Record the specific time when the medication reaction occurred. Associated drug combinations: The drugs or drug combinations that caused the medication reaction.

[0085] Pre-set medication rule library: Stores standardized labels and corresponding drug combinations for known medication reactions. The standardized labels use a unified medical terminology system (such as the WHO adverse reaction terminology library) to encode symptoms. For example, aspirin + ibuprofen → "gastrointestinal discomfort".

[0086] Pre-built patient comorbidity information database: Obtain the patient's currently active comorbidity types and their typical clinical manifestation labels, such as diabetes → "blood glucose fluctuations" and "polydipsia and polyuria".

[0087] For example, the symptoms of drug reaction are matched with standardized labels of known drug-associated reactions to determine the degree of matching. If the degree of matching reaches a preset threshold (e.g., 80%), it is determined to be a known reaction.

[0088] For example, for the reactions remaining after the first level of exclusion, their symptom labels are matched with the typical clinical manifestation labels of the patient's current active comorbidities. If the symptoms are highly similar to the clinical manifestations of the underlying disease (e.g., a match greater than 70%) and occur during the active phase of the disease, they are excluded.

[0089] It should be noted that the matching techniques used are well-known to those skilled in the art and will not be elaborated upon in this embodiment. For example, based on a medical terminology similarity algorithm, the semantic distance between the symptom description and the standard label is calculated and quantified as the matching degree.

[0090] S103: Risk fusion module, which communicates with the parameter calculation module and is used to determine the medication conflict risk value based on basic risk indicators, lag risk coefficients and group medication association risk parameters.

[0091] It is important to understand that many medication conflicts may not be obvious in individual cases, but they can be statistically significant at the population level. Therefore, by quantifying risk values, potential safety hazards can be identified in advance.

[0092] In this embodiment, the arithmetic mean of the basic risk indicators of all target drug combinations in the proposed medication regimen is calculated as the first risk indicator; the product of the first risk indicator and the lagged risk coefficient is calculated to obtain the superposition value of individual immediate and lagged risks; the superposition value is added to the group medication-related risk parameter to obtain the medication conflict risk value.

[0093] It is important to understand that the first risk indicator reflects the average baseline risk level of the drug combination in the proposed treatment regimen; the lagged risk coefficient reflects the risk that appears after a period of medication; the superposition value comprehensively reflects the immediate risk (conflict that occurs immediately after medication) and the lagged risk (conflict that occurs late); and the population-related risk parameter reflects the statistical risk level of the treatment regimen in the reference population. Therefore, if the first risk indicator, the lagged risk coefficient, and the population-related risk parameter are all larger, it indicates that the current treatment regimen not only has more significant medication conflicts on the basis of individual risk, but also has a higher risk at the population level.

[0094] S104: Prescription optimization module, which communicates with the risk fusion module. It is used to call the optimization algorithm to iteratively optimize the proposed medication plan with the goal of minimizing the risk value of medication conflict, and generate the optimal medication plan.

[0095] In this embodiment, the proposed medication regimen is used as the initial population of individuals, and the reciprocal of the drug conflict risk value is used as the fitness function. A new population is generated through selection, crossover, and mutation operations, where the mutation operation includes random perturbations of drug type, dosage, and administration time. The above process is iteratively executed until the convergence condition is met, and the individual with the highest fitness is finally output as the preferred medication regimen. Throughout the optimization process, all generated medication regimens must meet the pharmacologically safe dosage range and contraindications.

[0096] It should be noted that the reciprocal of the drug conflict risk value is used as the fitness function. The smaller the drug conflict risk value, the higher the fitness, ensuring that the optimization direction is to minimize the risk.

[0097] It should be noted that genetic algorithms are a well-known technique in the field of genetics, and the selection, crossover and mutation processes will not be described in detail in this embodiment.

[0098] It should be noted that after the iteration is completed, a risk comparison and adjustment explanation before and after optimization are also provided, such as "reducing the dosage of drug A from 5mg to 4mg reduces the drug conflict risk value by 0.23".

[0099] It should be noted that the convergence condition can be that the number of iterations reaches a preset value (such as 100 generations) or the decrease in the drug conflict risk value after 10 consecutive generations of optimization is less than 1%.

[0100] The pharmacologically safe dosage range can be dynamically determined based on the drug instructions and the individual patient's condition. There are no specific limits on the specific content. For example, the safe dosage range for adults of ibuprofen is 200–400 mg / dose, not exceeding 1200 mg per day.

[0101] Contraindications: Exclude medications that conflict with the patient's comorbidities.

[0102] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0103] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. An automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities, characterized in that, The system includes: The data management module is used to collect individual patient data and group data of the comorbidity reference group, and to receive the proposed medication plan. Individual data includes medication records, daily routine data and physiological indicators, while group data includes group medication records, medication response data and drug metabolism data. The parameter calculation module, which communicates with the data management module, is used to identify target drug combinations that overlap in time between patient medication records and proposed medication regimens; determine basic risk indicators based on the associated reaction data of target drug combinations in the group medication response data, patient medication records, and daily routine text data; determine the lag risk coefficient based on the temporal fluctuation characteristics of physiological indicators and drug metabolism data; and determine the group medication-related risk parameters based on the frequency of atypical reactions in the medication response data and the distribution characteristics of group medication records. The risk fusion module communicates with the parameter calculation module and is used to determine the medication conflict risk value based on basic risk indicators, lag risk coefficients and group medication association risk parameters. The prescription optimization module communicates with the risk fusion module and is used to call the optimization algorithm to iteratively optimize the proposed medication plan with the goal of minimizing the risk of medication conflict, so as to generate the optimal medication plan.

2. The automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities as described in claim 1, characterized in that, The identification of target drug combinations that overlap in time between the patient's medication records and the proposed medication regimen includes: Obtain the historical medication time series from the patient's medication record and the proposed medication time series from the proposed medication plan to construct a unified medication time axis. The historical medication time axis marks the actual medication time and the drug's duration of action, while the proposed medication time axis marks the planned medication time and the planned duration of action. Obtain the minimum safe dosing interval standard for each drug pair involved in the medication timeline; The effects of different drugs in the medication timeline are compared. If the overlap of the effects of any two drugs exceeds the safety threshold, or the planned medication interval is less than the corresponding minimum safe medication interval, then the two drugs are identified as target drugs and combined to form a target drug combination with overlapping medication times.

3. The automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities according to claim 2, characterized in that, The process for determining the basic risk indicators includes: The statistical data on adverse reactions in the population were analyzed, including the first person to experience an adverse reaction after using the target drug combination, the total number of people in the reference population who experienced adverse reactions after using the target drug alone; The ratio of the first group of patients to the total number of patients using the medication was calculated as the adverse reaction rate in the combination therapy group; the ratio of the second group of patients to the total number of patients using the medication was calculated as the adverse reaction rate in the single-drug group. For each group of target drug combinations, a baseline value for group risk is obtained by using a preset group risk calculation function based on the adverse reaction incidence rate of the combination group and the adverse reaction incidence rate of each target drug in the individual drug group. Based on patient medication records and daily routine text data, individual risk correction factors were determined; The product of the individual risk adjustment factor and the group risk benchmark is calculated as the basic risk indicator.

4. The automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities as described in claim 3, characterized in that, The process for determining the individual risk correction factor includes: Extract the time difference between two consecutive doses of the same drug from historical medication time series to form a time difference dataset; divide the time difference dataset into a predetermined number of continuous intervals, count the number of time differences in each interval and the total number of time differences; calculate the proportion of the number of time differences in each interval to the total number, as the interval time frequency value. The entropy value of all time frequency values ​​in all intervals is calculated using the entropy method. The natural logarithm of the total number of intervals is calculated as the limiting entropy value. The ratio of the entropy value to the limiting entropy value is normalized and used as the medication regularity score. Natural language processing is performed on the daily routine text data. Keywords related to daily routines are extracted through word segmentation. Positive and negative features are identified based on the medical word vector model, and the frequency of feature occurrence is statistically analyzed. The frequency of positive and negative features is input into a pre-trained daily routine scoring model, and a normalized daily routine score is output. The patient's medication adherence score and daily routine score are added together to obtain the total score; the ratio of the total score to the arithmetic mean of the total scores of all patients is calculated as an individual risk correction factor to quantify the differences in individual and group patterns.

5. The automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities according to claim 3, characterized in that, The process for determining the lag risk coefficient includes: Acquire the mean steady-state drug concentration and population metabolic characteristic coefficients from drug metabolism data; and extract the total number of days the target drug was taken from patient medication records; For each target drug, the product of the average steady-state drug concentration and the population metabolic characteristic coefficient is calculated as the baseline metabolic index; the ratio of the baseline metabolic index to the total number of days for the corresponding target drug is calculated as the residual impact coefficient reflecting the potential residual risk of the drug. For any target drug in the target drug combination, the incidence rate of adverse reactions in the single-drug group of the target drug is increased by a preset multiple as a population reference threshold; based on the incidence rate of adverse reactions in the combination group, the population reference threshold, and the residual effect coefficient, the probability of lagged risk is determined. Extract multiple time-series detection values ​​of the same physiological indicator within a preset time period and the preset normal range of the physiological indicator from the physiological indicators; normalize the ratio of the standard deviation of the time-series detection values ​​to the width of the preset normal range of the physiological indicator to obtain the physiological state fluctuation index. The sum of the positive integer 1 and the physiological state fluctuation index is calculated as the adjustment factor; the product of the lagged risk probability and the adjustment factor is calculated as the lagged risk coefficient.

6. The automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities according to claim 5, characterized in that, The determination of the lagged risk probability based on the adverse reaction incidence rate of the combined use group, the population reference threshold, and the residual effect coefficient includes: For each target drug combination, based on a preset combination weighting coefficient, the residual influence coefficient of the individual drug in the target drug combination is coupled into a combination residual influence coefficient; Select the maximum population reference threshold for a single drug from the target drug combination; The average difference between the incidence of adverse reactions in the combination group of the target drug combination and the incidence of adverse reactions in the single-drug group of the single target drug in the combination is calculated as the additional risk increment; Based on preset weighting coefficients, the maximum group reference threshold and the additional risk increment are weighted and fused to obtain the superimposed risk value; the product of the superimposed risk value and the combined residual influence coefficient is calculated as the lag risk probability.

7. The automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities according to claim 3, characterized in that, The process for determining the risk parameters associated with group medication use includes: Screen for atypical reactions in the group drug response data, count the number of target patients associated with the target drug combination of atypical reactions and the number of atypical reaction events, and calculate the ratio of the number of atypical reaction events to the number of target patients as the frequency of atypical reaction. The frequency of occurrence of any target drug combination in the medication records of the statistical reference group is calculated; the information entropy of all occurrence frequencies is normalized to obtain the distribution complexity index. The ratio of the frequency of atypical reactions to the distribution complexity index is calculated and used as a parameter for the risk associated with group medication use.

8. The automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities according to claim 7, characterized in that, The screening for atypical reactions is achieved through a two-stage exclusion mechanism, including: The system retrieves symptom tags, occurrence times, and associated drug combinations from population medication response data; it also calls a pre-set medication rule base to retrieve standardized tags and corresponding drug combinations for known medication responses; and it calls a pre-set patient comorbidity information base to retrieve the patient's currently active comorbidity types and their typical clinical manifestation tags. First-level exclusion: Match the drug reaction label with the standardized labels of known drug-associated reactions. If a match is found, it is identified as a known reaction and excluded. Second-level exclusion: For the unmatched medication responses remaining after the first-level exclusion, the remaining medication response tags are further matched with the typical clinical manifestation tags of the patient's currently active comorbidities. If a match is found, the response is excluded. The remaining drug reactions after the above two-stage exclusion are called atypical reactions and are stored in the atypical reaction dataset. At the same time, their frequency of occurrence in the population and information on associated drug combinations are recorded.

9. The automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities according to claim 1, characterized in that, The process for determining the risk value of medication conflicts includes: Calculate the arithmetic mean of the basic risk indicators of all target drug combinations in the proposed treatment plan, and use it as the first risk indicator; The product of the first risk indicator and the lagged risk coefficient is calculated to obtain the sum of the individual immediate and lagged risks; the sum is added to the group medication-related risk parameter to obtain the medication conflict risk value.

10. The automatic medication conflict detection and prescription optimization system for elderly patients with multiple comorbidities according to claim 1, characterized in that, The step of minimizing the risk of medication conflicts by calling an optimization algorithm to iteratively optimize the proposed medication regimen and generate an optimal medication regimen includes: The initial population of individuals is based on the proposed drug administration plan, and the fitness function is the reciprocal of the drug conflict risk value. New populations are generated through selection, crossover, and mutation operations, where mutation operations include random perturbations of drug type, dosage, and administration time. The above process is iteratively executed until the convergence condition is met, and the individual with the highest fitness is finally output as the preferred medication regimen. Throughout the optimization process, all generated medication regimens must meet the pharmacologically safe dosage range and contraindications.

Images