CN-121983230-A - Be used for methotrexate to give medicine prescription identification matching method

CN121983230ACN 121983230 ACN121983230 ACN 121983230ACN-121983230-A

Abstract

The invention discloses a prescription identification matching method for methotrexate administration, which particularly relates to the fields of intelligent prescriptions and medical informatics and is used for solving the problems that in the prior art, a methotrexate administration scheme mainly depends on doctor experience adjustment and is difficult to carry out individual matching by comprehensively utilizing historical case data, constructing a disease activity manifold track by extracting historical clinical records in an electronic medical record of a target patient, analyzing a delay accumulation rule of disease state change after administration by combining prescription records to obtain a drug effect attenuation state parameter, identifying a patient attenuation subtype on the basis, carrying out causality analysis to obtain a causal contribution weight vector, retrieving a similar patient group from a historical case library, generating a candidate intervention track set containing success probability distribution, and finally generating a methotrexate administration scheme recommendation list according to the order of success probability, thereby providing a methotrexate administration scheme with more reasonable and reference value for the target patient.

Inventors

Cai Jiaqin
SUN HONG
WEI XIAOXIA
HUANG YAPING
ZHANG JIAHAO

Assignees

福州大学附属省立医院

Dates

Publication Date: 20260505
Application Date: 20260401

Claims (7)

1. A method for identifying and matching prescriptions for methotrexate administration, comprising the steps of: S1, extracting a historical clinical record from an electronic medical record of a target patient, integrating recorded clinical indexes into time series data, and generating a manifold track of disease activity through dimension reduction processing; S2, extracting prescription records of a target patient, converting the dosage, interval and combined medication information of each methotrexate administration scheme into characteristic data, aligning the characteristic data with a disease activity manifold track according to time, and analyzing a delay accumulation rule of disease state change after medication to obtain a drug effect attenuation state parameter; s3, merging the efficacy attenuation state parameters and the healthy baseline of the target patient into attenuation feature vectors, and inputting an attenuation subtype identification model to identify an attenuation subtype label to which the target patient belongs; S4, carrying out causal relation analysis on clinical index data of a target patient and methotrexate dosing scheme characteristic data according to the attenuation subtype labels, and generating a causal contribution weight vector comprising a medication intensity factor contribution rate, a disease progression contribution rate and a complications interference contribution rate; s5, searching similar patient groups from a historical case library by taking the causal contribution weight vector as a similarity judgment basis, extracting methotrexate dosing regimen adjustment events and clinical indexes of similar patients, and generating a candidate intervention track set comprising multiple intervention branches and success probability distribution; s6, arranging each intervention branch in descending order according to the success probability of each candidate intervention branch, and generating a prescription recommendation list containing a plurality of methotrexate administration schemes according to the sorting result.
2. The method for identifying and matching a prescription for methotrexate administration according to claim 1, wherein in S1, generating a manifold trace of disease activity specifically comprises: Extracting joint tenderness count, swelling joint count, morning stiffness duration and health assessment questionnaire score recorded during each visit from historical clinical records of a target patient as clinical indexes, and arranging according to the visit time sequence to form a multi-mode clinical index time sequence matrix; In a multi-mode clinical index time sequence matrix, calculating the distance between any two time point vectors, and constructing a distance matrix reflecting the similarity between disease states; Performing feature decomposition on the distance matrix by adopting equidistant feature mapping, mapping each time point vector to a three-dimensional space, and obtaining a dimension-reduction coordinate sequence of each time point in the three-dimensional space, wherein coordinate points in the dimension-reduction coordinate sequence are used as disease activity degrees corresponding to a target patient; and carrying out cubic spline interpolation on the three-dimensional coordinates in the dimension-reduction coordinate sequence by taking the treatment time as a parameter variable to generate a disease activity manifold track reflecting the disease state evolution process of the target patient along with time.
3. The method for identifying and matching a prescription for methotrexate administration according to claim 1, wherein in S2, analyzing a delay accumulation rule of a disease state change after administration of the drug to obtain a drug effect attenuation state parameter specifically includes: extracting the administration dosage of each methotrexate prescription, the interval days of two adjacent administrations and the type of combined administration from the prescription record of a target patient, and constructing an administration characteristic vector sequence according to the prescription time; Matching and aligning the prescribing time in the administration characteristic vector sequence with a time axis of a disease activity manifold track, positioning a disease activity value corresponding to the prescribing time as a base line value, and intercepting a disease activity track fragment in a fixed time window after each prescription prescribing time as an administration effect sequence; Performing nonlinear least square fitting on the effect sequence of each administration by adopting a first-order exponential decay model, wherein the first-order exponential decay model takes a base line value as a starting point, takes time as an independent variable and takes a disease activity value as a dependent variable, and the initial descending amplitude and decay rate constant of the disease activity value after the administration are obtained by fitting; And calculating the residual effect superposition value of each administration at the current moment according to the initial drop amplitude and the attenuation rate constant of the disease activity value after the administration of the past times, and integrating the residual effect superposition value and the average attenuation rate constant of the past times as the drug effect attenuation state parameter.
4. The method for identifying and matching a prescription for methotrexate administration according to claim 1, wherein in S3, identifying the attenuation subtype tag to which the target patient belongs specifically comprises: Extracting healthy baseline data containing age, course and complications from an electronic medical record of a target patient, and splicing the healthy baseline data with a residual effect superposition value and an average attenuation deceleration rate constant in a drug effect attenuation state parameter to construct an attenuation characteristic vector; The attenuation subtype identification model selects a Gaussian mixture model, an attenuation characteristic vector is input into the Gaussian mixture model, the Gaussian mixture model consists of three Gaussian components, each Gaussian component corresponds to one attenuation subtype, the input is the attenuation characteristic vector, and the output is the posterior probability of the attenuation characteristic vector belonging to each Gaussian component; And selecting a Gaussian component corresponding to the maximum value from posterior probabilities output by the Gaussian mixture model, and taking an attenuation subtype tag pre-associated with the Gaussian component as an attenuation subtype tag of a target patient, wherein the attenuation subtype tag comprises insufficient medication intensity attenuation, progressive disease attenuation and complications interference attenuation.
5. The method of claim 4, wherein the gaussian mixture model extracts attenuation feature vectors of the historical patients labeled with attenuation subtype labels from the historical case library as a training set, and the mean vector and covariance matrix of each gaussian component are estimated by using a expectation maximization algorithm, and training is performed on the training set by using a log likelihood function maximization method.
6. The method of claim 1, wherein generating a causal contribution weight vector comprising a medication intensity factor contribution rate, a disease progression contribution rate, and a complications interference contribution rate in S4 specifically comprises: extracting multimode clinical indexes of a target patient in an attenuation subtype tag identification time window to serve as an endogenous observation variable, and extracting corresponding administration characteristic vectors to serve as exogenous observation variables; Constructing a structural equation model containing a medication intensity latent variable, a disease progression latent variable and a complications interference latent variable by taking an attenuation subtype tag as a grouping variable, wherein the medication intensity latent variable is measured by a medication characteristic vector, the disease progression latent variable is measured by a change trend of a multi-mode clinical index, and the complications interference latent variable is measured by a complications category and a disease course; substituting the endogenous observation variable and the exogenous observation variable into a structural equation model, and estimating the path coefficient from each latent variable to the drug effect attenuation state parameter by adopting a weighted least square method to obtain a first path coefficient corresponding to the drug intensity latent variable, a second path coefficient corresponding to the disease progression latent variable and a third path coefficient corresponding to the complications interference latent variable; And normalizing the first path coefficient, the second path coefficient and the third path coefficient to obtain causal contribution weight vectors respectively corresponding to the intensity factor contribution rate, the disease progression contribution rate and the complications interference contribution rate.
7. The method of claim 1, wherein in S5, generating a candidate intervention trajectory set comprising multiple intervention branches and a probability distribution of success comprises: extracting methotrexate administration dosage adjustment values, interval adjustment values, combined medication changes and clinical indexes corresponding to clinical records corresponding to the prescription records of similar patient groups to form individual intervention tracks of the similar patients; classifying individual intervention tracks of all similar patients, classifying the tracks with the same administration adjustment type into the same intervention branch, counting the number of similar patients in each intervention branch and the proportion of clinical index changes reaching preset clinical alleviation judging conditions, and calculating the success probability of the branch; Summarizing the adjusted dosing scheme and the corresponding success probability of each intervention branch dosing, and generating a candidate intervention track set containing multiple intervention branches and success probability distribution.

Description

Be used for methotrexate to give medicine prescription identification matching method Technical Field The invention relates to the technical fields of intelligent prescriptions and medical informatics, in particular to a prescription identification and matching method for methotrexate administration. Background Methotrexate is a basic antirheumatic drug widely applied to the treatment of autoimmune diseases such as rheumatoid arthritis, and in clinical practice, the methotrexate is usually used in a long-term maintenance treatment mode, and doctors need to continuously adjust the dosage, the interval and the combined medication scheme according to the disease activity change, the past medication response and the complications of patients so as to obtain better treatment effect. However, due to the significant individual differences in patients, there is a significant difference in the disease response of different patients after methotrexate administration, e.g., some patients may get better relief from symptoms during the initial treatment phase, while others may gradually experience diminished efficacy or sustained disease activity over time. In addition, the age, duration, complications and past treatment history of patients also have important effects on the therapeutic effects of the drugs, so that clinical administration decisions have high complexity. At present, a doctor mainly depends on experience and disease evaluation results of regular follow-up to adjust a dosing scheme, and personalized treatment can be realized to a certain extent by the mode, but due to lack of systematic analysis on historical clinical data and past treatment tracks of patients, it is often difficult to comprehensively utilize treatment experience in historical cases, and it is also difficult to effectively compare and match treatment modes among different patients. Therefore, how to use multi-source clinical data in electronic medical records to establish the association relationship between the disease state change of a patient and the drug treatment response, and provide more reasonable methotrexate dosing scheme recommendation for a target patient on the basis, becomes a technical problem to be solved. Disclosure of Invention In order to overcome the above-mentioned drawbacks of the prior art, embodiments of the present invention provide a prescription identification matching method for methotrexate administration to solve the above-mentioned problems set forth in the background art. In order to achieve the above purpose, the present invention provides the following technical solutions: a method of identifying and matching a prescription for methotrexate administration comprising the steps of: S1, extracting a historical clinical record from an electronic medical record of a target patient, integrating recorded clinical indexes into time series data, and generating a manifold track of disease activity through dimension reduction processing; S2, extracting prescription records of a target patient, converting the dosage, interval and combined medication information of each methotrexate administration scheme into characteristic data, aligning the characteristic data with a disease activity manifold track according to time, and analyzing a delay accumulation rule of disease state change after medication to obtain a drug effect attenuation state parameter; s3, merging the efficacy attenuation state parameters and the healthy baseline of the target patient into attenuation feature vectors, and inputting an attenuation subtype identification model to identify an attenuation subtype label to which the target patient belongs; S4, carrying out causal relation analysis on clinical index data of a target patient and methotrexate dosing scheme characteristic data according to the attenuation subtype labels, and generating a causal contribution weight vector comprising a medication intensity factor contribution rate, a disease progression contribution rate and a complications interference contribution rate; s5, searching similar patient groups from a historical case library by taking the causal contribution weight vector as a similarity judgment basis, extracting methotrexate dosing regimen adjustment events and clinical indexes of similar patients, and generating a candidate intervention track set comprising multiple intervention branches and success probability distribution; s6, arranging each intervention branch in descending order according to the success probability of each candidate intervention branch, and generating a prescription recommendation list containing a plurality of methotrexate administration schemes according to the sorting result. As a further aspect of the present invention, in S1, generating a manifold track of disease activity specifically includes: Extracting joint tenderness count, swelling joint count, morning stiffness duration and health assessment questionnaire score recorded during each visit from hist