A tumor patient drug side reaction risk prediction system based on multi-source data

Abstract

The present application relates to the technical field of drug side reaction risk prediction, and particularly relates to a tumor patient drug side reaction risk prediction system based on multi-source data. The system comprises: a data acquisition module for integrating multi-source time series data, including drug dosage, pharmacokinetic parameters and clinical index changes, and constructing a multi-dimensional data set; a cumulative effect analysis module for calculating a drug attenuation factor based on first-order elimination kinetics, quantifying the cumulative effect intensity in combination with the dosage and median lethal dose, and capturing the drug metabolism dynamics; since there may be interactive effects between drugs, an influence coefficient analysis module is used to analyze the drug main effect and the joint drug interaction effect by using elastic network regression, and to determine the main influence coefficient and the interaction influence coefficient of the drug on each clinical index; a risk prediction module for optimizing the decision tree sorting by the abnormal contribution degree weighting, and outputting the risk probability of each side reaction in combination with the random forest algorithm.

Description

Technical Field

[0001] This invention relates to the field of drug side effect risk prediction technology, and specifically to a drug side effect risk prediction system for cancer patients based on multi-source data. Background Technology

[0002] With advancements in medical technology, drug therapy is now a common treatment option for cancer patients. There are numerous types of cancer drugs, which can be broadly categorized into five classes based on their mechanisms of action: chemotherapy drugs, targeted therapies, immunotherapy drugs, hormonal drugs, and adjuvant drugs. However, drug therapy can also lead to adverse side effects, such as bone marrow suppression, gastrointestinal reactions, cardiotoxicity, hepatotoxicity, pulmonary toxicity, neurotoxicity, urinary tract toxicity, and reproductive toxicity. Each side effect corresponds to fluctuations in several clinical indicators; therefore, monitoring these clinical indicators can help predict the risk of various side effects.

[0003] Existing technologies typically rely on random forest algorithms to predict the risk of side effects. However, since cancer patients often receive combined treatment with multiple drugs, there may be interactions between different drugs, and drugs can also have cumulative effects in the patient's body. Therefore, if decision trees are directly constructed based on basic data such as the dosage of various drugs, the resulting risk prediction indicators will have significant errors. Summary of the Invention

[0004] To address the challenges of combining multiple medications for cancer patients, which can lead to interactions and cumulative effects, directly constructing decision trees based on drug dosages and other fundamental data can result in significant errors in the generated risk prediction indicators. This invention aims to provide a cancer patient drug side effect risk prediction system based on multi-source data. The specific technical solution adopted is as follows:

[0005] This invention proposes a drug side effect risk prediction system for cancer patients based on multi-source data, the system comprising:

[0006] The data acquisition module is used to acquire the dosage, elimination rate constant, median lethal dose, time-series data of changes in various clinical indicators after each medication administration, and duration of drug action for each patient.

[0007] The cumulative effect analysis module is used to obtain the decay factor of each drug based on the first-order elimination kinetic principle, according to the elimination rate constant and drug duration of each drug; and to obtain the cumulative effect intensity of each drug based on the dose, median lethal dose and decay factor of each drug over all doses.

[0008] The impact coefficient analysis module is used to establish an elastic network regression model based on the cumulative effect intensity of the drug, the preset impact weight of the drug on clinical indicators, and the difference characteristics between the time series data of each clinical indicator after drug administration and the preset comparison time series data, to determine the impact coefficient of the drug on each clinical indicator. The impact coefficient includes the main impact coefficient and the interaction impact coefficient.

[0009] The risk prediction module is used to update the decision order of the decision tree by using the influence coefficient of the drug, combined with the cumulative effect intensity of the drug and the preset influence weight, so as to predict the risk probability of each side effect.

[0010] Furthermore, based on the principle of first-order elimination kinetics, the attenuation factor of each drug is obtained according to its elimination rate constant and duration of action, including:

[0011] The absolute value of the time difference between two consecutive medication times is taken as the duration of drug action at the time of the previous medication.

[0012] The attenuation factor of each drug is obtained by taking the negative of the product of the duration of drug action at each administration and the elimination rate constant of each drug, and using the natural constant as the base.

[0013] Furthermore, the method for obtaining the cumulative effect intensity includes:

[0014] After each administration, the ratio of the dose of each drug to the median lethal dose was used as the normalized dose, and the product of the normalized dose of each drug and the attenuation factor was used as the cumulative effect factor.

[0015] The normalized value of the cumulative effect factor corresponding to each drug after all administrations is taken as the cumulative effect intensity of each drug.

[0016] Furthermore, based on the cumulative effect intensity of the drug, the preset influence weights of the drug on clinical indicators, and the differences between the time-series data of each clinical indicator after drug administration and the preset comparison time-series data, an elastic network regression model is established to determine the influence coefficient of the drug on each clinical indicator. The influence coefficient includes the main influence coefficient and the interaction influence coefficient, including:

[0017] Analyze the differences between the time-series data of each clinical indicator after medication and the preset comparison time-series data to determine the indicator offset of each clinical indicator;

[0018] Combine any two drugs in pairs to obtain all unique drug combinations;

[0019] A regression function is constructed based on the cumulative effect intensity of drugs, the cumulative effect intensity of drugs in drug combinations, the preset influence weight of drugs on each clinical indicator, the preset main influence coefficient of each drug, and the preset interaction influence coefficient of drug combinations. The preset main influence coefficient and the preset interaction influence coefficient are variable, and the preset main influence coefficient is not 0.

[0020] In the elastic network regression model, the value of the regression function with the smallest absolute value of the difference between the index offset and each clinical indicator is taken as the objective function value.

[0021] The preset main influence coefficient and preset interaction influence coefficient of each drug under the objective function value are used as the main influence coefficient of each drug and the interaction influence coefficient of the drug combination to which it belongs.

[0022] Furthermore, the method for obtaining the indicator offset includes:

[0023] For each clinical indicator, the difference between the mean value of the medication time series data after each medication administration and the mean value of the preset comparison time series data is calculated, which is used as the indicator shift factor for each clinical indicator after each medication administration.

[0024] The mean of the index offset factor for each clinical indicator across all drug administration times is used as the index offset for each clinical indicator.

[0025] Furthermore, the formula model of the regression function includes:

[0026]

[0027] in, This represents the value of the j-th clinical indicator under the regression function; This represents the preset influence weight of the i-th drug on the j-th clinical indicator; This represents the pre-defined main influence coefficient of the i-th drug on the j-th clinical indicator; This represents the cumulative effect intensity of the i-th drug; n represents the number of drug types; m represents the total number of drug combinations. This represents the interaction coefficient between the two drugs in the k-th drug combination; This represents the cumulative effect intensity of drug 1 in the k-th drug combination; This represents the cumulative effect intensity of drug 2 in the k-th drug combination; This is the random error term.

[0028] Furthermore, the step of updating the decision order of the decision tree by utilizing the influence coefficient corresponding to the drug, combined with the cumulative effect intensity of the drug and a preset influence weight, to predict the risk probability of each side effect includes:

[0029] The abnormal contribution of each drug to each clinical indicator is determined based on the main influence coefficient of each drug on each clinical indicator and the interaction influence coefficient of the drug combination to which it belongs.

[0030] Determine the corresponding clinical indicators for each type of adverse reaction;

[0031] Based on the random forest algorithm, for any side reaction, several decision trees are constructed. In each decision tree, the decision order of the decision tree is updated by the descending order of the abnormal contribution, thereby outputting the risk factor of that side reaction.

[0032] The risk probability of this type of side reaction is obtained by normalizing the mean of the risk factors under all decision trees.

[0033] Furthermore, the method for obtaining the abnormal contribution degree includes:

[0034] Choose one drug as the test drug and choose one clinical indicator as the test indicator.

[0035] Under the target indicator, the product of the absolute value of the main influence coefficient of the test drug, the cumulative effect intensity of the test drug, and the preset influence weight of the test drug on the target indicator is used as the first abnormal contribution factor of the test drug to the target indicator.

[0036] Under the test index, when the interaction coefficient of all drug combinations to which the test drug belongs is 0, the normalized value of the first abnormal contribution factor of the test drug to the test index is taken as the abnormal contribution degree of the test drug to the test index.

[0037] Under the target indicator, when the interaction coefficients of all drug combinations to which the target drug belongs are not all zero, the drug combinations with non-zero interaction coefficients are taken as the target combinations. In each target combination, the proportion of the cumulative effect intensity of the target drug is analyzed. Then, combined with the interaction coefficients and the preset influence weights of the target drug on the target indicator, the second abnormal contribution factor of the target drug on the target indicator is obtained. The normalized value of the sum of the first and second abnormal contribution factors of the target drug on the target indicator is taken as the abnormal contribution degree of the target drug on the target indicator.

[0038] Furthermore, the method for obtaining the second abnormal contribution factor includes:

[0039] Under the test index, in each test combination, the proportion of the cumulative effect intensity of the test drug is used as the scaling factor, and the product of the scaling factor and the absolute value of the interaction coefficient corresponding to the test combination is used as the adjustment factor.

[0040] The sum of the adjustment factors of all test combinations and the product of the preset influence weight of the test drug on the test indicator are used as the second abnormal contribution factor of the test drug to the test indicator.

[0041] Furthermore, the preset comparison time series data consists of time series data on the changes of various clinical indicators under the condition of no medication.

[0042] The present invention has the following beneficial effects:

[0043] In the data acquisition module, a multi-dimensional dataset was constructed by integrating the dosage, elimination rate constant, median lethal dose (LD50), and time-series data of changes in each clinical indicator after each patient's medication administration. This provides a data foundation for subsequent accurate risk prediction. Given the dynamic cumulative effect of drugs in the patient's body, the cumulative effect analysis module quantifies the drug decay factor based on the first-order elimination kinetic model, considering the elimination rate constant and duration of drug action for each drug. This, combined with the drug dosage and LD50, calculates the cumulative effect intensity of the drug, thereby capturing the drug metabolism patterns and dynamic cumulative changes. Since cancer patients often receive combined treatment with multiple drugs, and the toxicity of multiple drugs may amplify or offset the impact on certain clinical indicators when multiple drugs are present simultaneously, the influence coefficient analysis module establishes a multi-dimensional dose-effect mapping relationship. Using an elastic network regression algorithm, based on the cumulative effect intensity of the drug, the differences between the time-series data of each clinical indicator after medication administration and preset comparison time-series data, an elastic network regression model is established to determine the main influence coefficient and interaction influence coefficient of the drug on each clinical indicator. Finally, in the risk prediction module, the decision order of the decision tree is determined by using the influence coefficient of the drug and combining it with the cumulative effect intensity of the drug, so as to improve the prediction accuracy of the random forest model and obtain the risk probability of each side reaction. Attached Figure Description

[0044] 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.

[0045] Figure 1 This is a system block diagram of a tumor patient drug side effect risk prediction system based on multi-source data, provided in one embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the system structure of a tumor patient drug side effect risk prediction system based on multi-source data, provided in one embodiment of the present invention. Detailed Implementation

[0047] 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 a multi-source data-based drug side effect risk prediction system for cancer patients 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.

[0048] 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.

[0049] The following description, in conjunction with the accompanying drawings, details a specific scheme for a drug side effect risk prediction system for tumor patients based on multi-source data, provided by the present invention.

[0050] Please see Figure 1 The diagram illustrates a system block diagram of a tumor patient drug side effect risk prediction system based on multi-source data according to an embodiment of the present invention. The system includes: a data acquisition module 101, a cumulative effect analysis module 102, an influence coefficient analysis module 103, and a risk prediction module 104.

[0051] The data acquisition module 101 is used to acquire the dosage of each drug, elimination rate constant, median lethal dose, time-series data of changes in various clinical indicators after each drug administration, and duration of drug action for each patient.

[0052] Drug therapy can be used to treat cancer patients; however, drug therapy can also cause adverse side effects, such as bone marrow suppression, gastrointestinal reactions, cardiotoxicity, hepatotoxicity, pulmonary toxicity, neurotoxicity, urinary tract toxicity, and reproductive toxicity. The side effects caused by anticancer drugs are specifically pathophysiologically related to changes in clinical indicators, and changes in different categories of indicators reflect different types of organ damage mechanisms.

[0053] Therefore, when analyzing drug side effects in cancer patients, multi-source data can be obtained, specifically including the dosage of each drug at each time the patient takes medication (which can be obtained through medication records in electronic medical records). After each medication, various clinical indicators of the patient are monitored to obtain time-series data of changes, and the duration of drug action at each medication is recorded. The duration of drug action can be recorded as the time interval between two consecutive medications. The length of the time-series data of changes in clinical indicators is set as the duration of drug action for each medication, and the sampling frequency can be set to once per second.

[0054] Based on prior knowledge, side effects after a patient takes anticancer drugs do not appear immediately. They mainly occur due to the accumulation of drug toxicity in the body. The median lethal dose (LD50) of a drug can provide a baseline for drug toxicity and provide dose standardization for subsequent analysis of side effect risk. Therefore, the LD50 of each drug can be obtained (through drug instructions, etc.). The elimination rate constant reflects the rate of drug metabolism in the body and can be derived from the half-life of each drug and the first-order elimination kinetics. The formula is as follows: ,in, Represents the elimination rate constant of the i-th drug; This represents the half-life of the i-th drug; This represents a logarithmic function with the natural constant e as its base.

[0055] It should be noted that, in this embodiment of the invention, the duration of drug action and the sampling frequency of time-series data can be adjusted according to the implementation scenario, and are not limited here; the elimination rate constant is obtained by formula derivation based on the first-order elimination kinetics principle and the drug half-life, which is a well-known content, and the process will not be described in detail here; in this embodiment of the invention, the types of clinical indicators can be selected from white blood cell count, platelet count, hemoglobin, etc., and the number of types can be set to 15. The specific types and numbers can be selected according to the implementation scenario, and are not limited here.

[0056] In this embodiment of the invention, the collection and acquisition of patient personal data are authorized by the relevant users, and the process does not violate relevant laws and regulations, nor does it violate public order and good morals.

[0057] The cumulative effect analysis module 102 is used to obtain the attenuation factor of each drug based on the first-order elimination kinetic principle, according to the elimination rate constant and drug duration of each drug; and to obtain the cumulative effect intensity of each drug based on the dose, median lethal dose and attenuation factor of each drug at all doses.

[0058] When clinical indicators deviate from the baseline level, they do not necessarily have obvious side effects. However, after long-term use of anticancer drugs, the toxicity of various drugs accumulates, eventually leading to side effects. Therefore, in the cumulative effect analysis module, based on the first-order elimination kinetic model, the drug attenuation factor is quantified according to the elimination rate constant of each drug and the duration of drug action. The attenuation factor can be used to characterize the proportion of drug concentration retained after a certain period of time. Then, this indicator is combined with the drug dose and median lethal dose to calculate the cumulative effect intensity of the drug, thereby obtaining the cumulative effect intensity of each drug and capturing the drug metabolism law and dynamic cumulative changes.

[0059] Preferably, in one embodiment of the present invention, the method for obtaining the attenuation factor includes:

[0060] The absolute value of the time difference between two consecutive medication administration times is taken as the duration of drug action at the time of the previous administration.

[0061] The attenuation factor for each drug is obtained by taking the negative of the product of the duration of drug action at each administration and the elimination rate constant of each drug, and using the natural constant as the base. The specific model for the attenuation factor is as follows: ,in, Represents the natural constant; Represents the elimination rate constant of the i-th drug; Indicates the duration of drug action.

[0062] The duration of drug action determines the decay duration, the elimination rate constant determines the decay rate, and the decay factor combines the two to determine the proportion of drug residue, which can be used to calculate the cumulative effect of toxic substances.

[0063] It should be noted that the formula for the attenuation factor can also be derived based on the first-order elimination kinetics principle. The specific derivation process is a well-known process and will not be elaborated here.

[0064] In pharmacological and toxicological analysis, the traditional Hill equation was originally used to describe the concentration-response relationship after a single dose, but it cannot characterize the dynamic process of drug metabolism. In order to consider the cumulative effect of multiple doses, in this embodiment of the present invention, the cumulative effect intensity of each drug is calculated by modifying the original Hill equation based on the dose, median lethal dose and decay factor of each drug for all doses.

[0065] Preferably, in one embodiment of the present invention, the method for obtaining the cumulative effect intensity includes:

[0066] After each administration, the ratio of the dose of each drug to the median lethal dose (LD50) was used as the normalized dose. The LD50 is the dose that causes 50% of the subjects to die, serving as the toxicity benchmark. The normalized dose converts the actual dose of each drug into relative toxicity units. The attenuation factor of each drug reflects the elimination process of the drug in the body. Therefore, the normalized dose of each drug was combined with the attenuation factor, and the product of the normalized dose and the attenuation factor was used as the cumulative effect factor. The larger the cumulative effect factor, the greater the drug toxicity residue.

[0067] Finally, the mean of the cumulative effect factor corresponding to each drug after all administrations is normalized and used as the cumulative effect strength of each drug. The higher the cumulative effect strength, the higher the cumulative toxicity of the drug in the patient. Normalization is a technique well known to those skilled in the art, and the normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0068] Thus, in the cumulative effect analysis module, the cumulative toxicity of each drug in the patient's body after several doses can be analyzed, and this can be used in the subsequent prediction of adverse reactions.

[0069] The influence coefficient analysis module 103 is used to establish an elastic network regression model based on the cumulative effect intensity of the drug, the preset influence weight of the drug on clinical indicators, and the difference characteristics between the time series data of each clinical indicator after drug use and the preset comparison time series data, to determine the influence coefficient of the drug on each clinical indicator. The influence coefficient includes the main influence coefficient and the interaction influence coefficient.

[0070] When patients take anticancer drugs, they may experience various side effects. Specific anticancer drug-induced side effects and changes in clinical indicators have specific pathophysiological associations, and changes in different categories of indicators reflect different types of organ damage mechanisms. That is, for any type of anticancer drug taken, specific side effects will occur, corresponding to specific changes in clinical indicators. Therefore, for each clinical indicator, the differences between the time-series data of changes after medication and the preset comparison time-series data (referring to the time-series data of changes in various clinical indicators in the state where the patient has not taken medication, i.e., the data duration can be set to 30 minutes) can be analyzed to characterize the fluctuation and deviation of clinical indicators after medication. Simultaneously, since patients take multiple drugs, and multiple drugs are present simultaneously, their toxicity load product may amplify or cancel the effects on certain indicators, i.e., synergistic / antagonistic effects occur. For example, the combined use of cisplatin and gemcitabine can exacerbate thrombocytopenia. Therefore, the effects of drugs on clinical indicators can be divided into primary effects and interactive effects.

[0071] To distinguish between the direct contribution and interaction effects of drugs on each clinical indicator, a multi-dimensional effect mapping relationship can be established. Specifically, an elastic network regression algorithm is used. Based on the cumulative effect intensity of the drug, the preset influence weight of the drug on the clinical indicator, and the difference characteristics between the time series data of each clinical indicator after drug administration and the preset comparison time series data, an elastic network regression model is constructed. The algorithm automatically identifies the main effects (main influence coefficient) and interaction effects (interaction influence coefficient) with statistical significance. At the same time, by adjusting the L1 / L2 regularization strength, it prevents excessive attention to occasional minor effects. Thus, the main influence coefficient of each drug on each clinical indicator and the interaction influence coefficient when drugs act together are analyzed.

[0072] Preferably, in one embodiment of the present invention, an elastic network regression model is established based on the cumulative effect intensity of the drug, the preset influence weight of the drug on the indicator, and the difference characteristics between the time-series data of each clinical indicator after drug administration and the preset comparison time-series data, to determine the influence coefficient of the drug on each clinical indicator. The influence coefficient includes the primary influence coefficient and the interaction influence coefficient of the drug combination, including:

[0073] First, the differences between the time-series data of each clinical indicator after medication and the preset comparison time-series data were analyzed to determine the indicator offset of each clinical indicator. For each clinical indicator, the mean value of the time-series data after each medication administration reflects the average characteristics of the clinical indicator after medication, which represents the average level under the influence of the drug. The mean value of the preset comparison time-series data serves as a baseline data, which is not affected by the drug. Therefore, the degree of deviation between the two is analyzed, and the difference between the average level after medication and the average level without drug influence is calculated as the indicator offset factor of each clinical indicator after each medication administration. The larger the absolute value of this value, the greater the degree of deviation of the clinical indicator under the influence of the drug, that is, the greater the influence. A positive value indicates that the clinical indicator is too high, and vice versa. The degree of deviation of each clinical indicator under all medication administration times is integrated, that is, the mean value of the indicator offset factor of each clinical indicator under all medication administration times is used as the indicator offset of each clinical indicator.

[0074] Based on drug instructions or literature evidence, and combined with the index offset of clinical indicators, the positive / negative effects of each drug on each clinical indicator are labeled as preset influence weights. For example, if the drug instructions or literature evidence states that the diuretic hydrochlorothiazide will cause a decrease in serum potassium, and the index offset of serum potassium is negative, it means that the patient's serum potassium decreased during actual medication. In this case, the efficacy of the diuretic hydrochlorothiazide matches the actual effect, and the preset influence weight of hydrochlorothiazide on serum potassium is +1, while the preset influence weight of potassium supplements on serum potassium is -1. This allows us to set the preset influence weight of each drug on each clinical indicator.

[0075] Since different types of drugs can interact with each other, any two drugs are combined to obtain all unique drug combinations. Then, a regression function is constructed based on the cumulative effect strength of the drugs, the cumulative effect strength of the drugs in the drug combination, the preset influence weight of each drug on each clinical indicator, the preset main influence coefficient of each drug, and the preset interaction influence coefficient of the drug combination. The preset main influence coefficient represents the independent effect of the drugs on the clinical indicators, and the interaction influence coefficient represents the interactive effect of the drugs on the clinical indicators under the combined action. Both are variable and can be adjusted through model optimization. In this embodiment of the invention, it is assumed that the preset main influence coefficient cannot be 0.

[0076] The formula model for the regression function includes:

[0077]

[0078] in, This represents the value of the j-th clinical indicator under the regression function; This represents the preset influence weight of the i-th drug on the j-th clinical indicator; This represents the pre-defined main influence coefficient of the i-th drug on the j-th clinical indicator; This represents the cumulative effect intensity of the i-th drug; n represents the number of drug types; m represents the total number of drug combinations. This represents the interaction coefficient between the two drugs in the k-th drug combination; This represents the cumulative effect intensity of drug 1 in the k-th drug combination; This represents the cumulative effect intensity of drug 2 in the k-th drug combination; represents the random error term, and represents the formula terms included in the algorithm.

[0079] In the regression function, It can be considered as the main effect term, used to capture the independent effects of the drug. The interaction term is used to capture the combined effects between drugs. Through L1 and L2 regularization constraints, the values ​​of each clinical indicator under the above regression function formula model are obtained, reflecting the predicted fluctuations of clinical indicators under the single effect and interaction effect of drugs.

[0080] Then, under the elastic network regression model, through grid search, the value of the regression function with the smallest absolute value of the difference between the regression function and the index offset (actual fluctuation of the clinical index) of each clinical index is taken as the objective function value.

[0081] Finally, the preset main influence coefficient and preset interaction influence coefficient of each drug under the objective function value are used as the main influence coefficient of each drug and the interaction influence coefficient of the drug combination to which it belongs.

[0082] Thus, we can obtain the main influence coefficient of each drug on each clinical indicator and the interaction influence coefficient under the combined action of drugs. For example, for a certain clinical indicator, the main influence coefficient of insulin is 0.35, while the interaction influence coefficient of metformin and insulin is 0.12.

[0083] It should be noted that the elastic network regression algorithm is a well-known technique, and the process will not be described in detail here.

[0084] The risk prediction module 104 is used to update the decision order of the decision tree by using the influence coefficient of the drug, combined with the cumulative effect intensity of the drug and the preset influence weight, so as to predict the risk probability of each side effect.

[0085] Based on prior knowledge, when a certain clinical indicator deviates from its baseline level, it does not necessarily indicate obvious side effects. However, with long-term use of anticancer drugs, the toxicity of various drugs accumulates, eventually leading to side effects. Therefore, it is necessary to continuously monitor various clinical indicators of patients during medication. At the same time, combining the random forest algorithm can output the risk probability of various side effects. However, directly constructing a basic random forest algorithm, inputting the dosage of various drugs and the corresponding deviation of various clinical indicators, and ignoring the cumulative effect of drug toxicity and the interaction effect between drugs, can easily lead to a large error in the generated risk probability of side effects.

[0086] Therefore, in this embodiment of the invention, in the risk prediction module, when constructing the decision tree, the decision order of the decision tree is updated by using the influence coefficient corresponding to the drug, combined with the cumulative effect intensity of the drug and the preset influence weight, so as to avoid large errors in the final prediction probability caused by splitting according to random nodes.

[0087] Preferably, in one embodiment of the present invention, the decision order of the decision tree is updated by utilizing the influence coefficient corresponding to the drug, combined with the cumulative effect intensity of the drug and a preset influence weight, thereby predicting the risk probability of each side effect, including:

[0088] Because drugs have both independent and combined effects on clinical indicators, the abnormal contribution of each drug to each clinical indicator can be determined based on the main influence coefficient of each drug on each clinical indicator and the interaction influence coefficient of the drug combination to which it belongs.

[0089] For ease of subsequent explanation and clarification, we will select one drug as the test drug and one clinical indicator as the test indicator.

[0090] We can first analyze the abnormal contribution of the test drug to the test indicator under independent action. The abnormal contribution is mainly determined by the main influence coefficient and the degree of toxicity accumulation. Under the test indicator, the product of the absolute value of the main influence coefficient of the test drug, the cumulative effect intensity of the test drug, and the preset influence weight of the test drug on the test indicator is used as the first abnormal contribution factor of the test drug on the test indicator. The sign of the first abnormal contribution factor is determined by the preset influence weight of the test drug on the test indicator. A negative sign indicates a negative effect, and a positive sign indicates a positive effect. The cumulative effect intensity of the test drug reflects the intensity of the cumulative toxicity effect. Therefore, the larger the value of the first abnormal contribution factor, the greater the abnormal contribution of the test drug to the test indicator under independent action, and the higher its priority.

[0091] Then, it is necessary to perform a fusion analysis on the first abnormal contribution factor under the toxicological effects of the test drug and the contribution of the test drug under the combined effects of other drugs to obtain the abnormal contribution degree of the test drug to the test index.

[0092] Under the test index, when the interaction coefficient of all drug combinations to which the test drug belongs is 0, it indicates that the test drug does not have a common influence with other drugs. Therefore, the normalized value of the first abnormal contribution factor of the test drug to the test index is directly used as the abnormal contribution degree of the test drug to the test index.

[0093] If, under the target indicator, the interaction coefficients of all drug combinations to which the target drug belongs are not all zero, it indicates that the target drug and other drugs have a common effect on the target indicator. Therefore, the drug combinations to which the target drug belongs with non-zero interaction coefficients are taken as the target combinations.

[0094] In each test combination, the influence is distributed according to the cumulative effect intensity of the drug. Therefore, under the test index, the proportion of the cumulative effect intensity of the test drug in the test combination is used as the proportional factor, and the product of the proportional factor and the absolute value of the interaction coefficient corresponding to the test combination is used as the adjustment factor. The adjustment factor reflects the degree of abnormal contribution of the test drug to the test index in the test combination.

[0095] At this point, the drug under test has an adjustment factor in each test combination. Therefore, the sum of the adjustment factors of all test combinations and the product of the drug under test's preset influence weight on the test indicator are used as the second abnormal contribution factor of the drug under test on the test indicator. Similarly, the sign of the second abnormal contribution factor is determined by the drug under test's preset influence weight on the test indicator. A negative sign indicates a negative influence, and a positive sign indicates a positive influence. The larger the value of the second abnormal contribution factor, the greater the abnormal contribution of the drug under test to the test indicator, and the higher its priority.

[0096] The normalized sum of the first and second abnormal contribution factors of the drug under test to the target indicator is taken as the abnormal contribution degree of the drug under test to the target indicator.

[0097] The clinical indicators corresponding to each type of adverse reaction can be determined based on prior knowledge. For example, for adverse reactions to the urinary system, the corresponding clinical indicators mainly include creatinine and cystatin.

[0098] Based on the random forest algorithm, for any side effect, several decision trees are constructed. In each decision tree, the index offset of the clinical indicators corresponding to the side effect, the type and dosage of the drug taken by the patient are used as input features. The decision order of the decision tree is updated by the descending order of the abnormal contribution, and the features with high abnormal contribution are segmented first, so as to output the risk factor of the side effect.

[0099] At this point, each side reaction has a risk factor under its corresponding decision tree; the larger the value, the greater the probability of that side reaction occurring. Finally, the mean of the risk factors for that side reaction under all decision trees is normalized to obtain the risk probability of that side reaction. Normalization is a technique well-known to those skilled in the art, and the choice of normalization function can be linear normalization or standard normalization, etc. The specific normalization method is not limited here.

[0100] It should be noted that the random forest algorithm is a well-known technique, and the specific process will not be described in detail here.

[0101] In summary, the data acquisition module integrates data on the dosage, elimination rate constant, median lethal dose (LD50), and time-series data on changes in each clinical indicator after each patient's medication administration, constructing a multi-dimensional dataset to provide a data foundation for subsequent accurate risk prediction. Given the dynamic cumulative effect of drugs in the patient's body, the cumulative effect analysis module quantifies the drug decay factor based on the first-order elimination kinetic model, considering the elimination rate constant and duration of drug action for each drug. This is combined with the drug dosage and LD50 to calculate the cumulative effect intensity, thereby capturing the drug metabolism patterns and dynamic cumulative changes. Since cancer patients often receive combined treatment with multiple drugs, and the toxicity of multiple drugs may amplify or offset the effects on certain clinical indicators when multiple drugs are present simultaneously, the influence coefficient analysis module establishes a multi-dimensional dose-effect mapping relationship. Using an elastic network regression algorithm, based on the cumulative effect intensity of the drug, the differences between the time-series data of each clinical indicator after medication administration and preset comparison time-series data, an elastic network regression model is established to determine the main influence coefficient and interaction influence coefficient of the drug on each clinical indicator. Finally, in the risk prediction module, the decision order of the decision tree is determined by using the influence coefficient of the drug and combining it with the cumulative effect intensity of the drug, so as to improve the prediction accuracy of the random forest model and obtain the risk probability of each side reaction.

[0102] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above.

[0103] Please see Figure 2 This illustration shows a schematic diagram of the system structure of a tumor patient drug side effect risk prediction system based on multi-source data according to an embodiment of the present invention. It includes a processor 200, a memory 201, a bus 202, and a communication interface 203. The processor 200, communication interface 203, and memory 201 are connected via the bus 202. The memory 201 may include a high-speed random access memory, and the bus 202 may be an ISA bus, PCI bus, or EISA bus, etc. The processor 200 may be an integrated circuit chip with signal processing capabilities. The memory 201 stores at least one instruction, at least one program, code set, or instruction set. When the processor loads and executes the at least one instruction, at least one program, code set, or instruction set, it implements the steps of a tumor patient drug side effect risk prediction system based on multi-source data.

[0104] 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.

[0105] 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.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A drug side effect risk prediction system for cancer patients based on multi-source data, characterized in that, The system includes: The data acquisition module is used to acquire the dosage, elimination rate constant, median lethal dose, time-series data of changes in various clinical indicators after each medication administration, and duration of drug action for each patient. The cumulative effect analysis module is used to obtain the decay factor of each drug based on the first-order elimination kinetic principle, according to the elimination rate constant and drug duration of each drug; and to obtain the cumulative effect intensity of each drug based on the dose, median lethal dose and decay factor of each drug over all doses. The impact coefficient analysis module is used to establish an elastic network regression model based on the cumulative effect intensity of the drug, the preset impact weight of the drug on clinical indicators, and the difference characteristics between the time series data of each clinical indicator after drug administration and the preset comparison time series data, to determine the impact coefficient of the drug on each clinical indicator. The impact coefficient includes the main impact coefficient and the interaction impact coefficient. The risk prediction module is used to update the decision order of the decision tree by using the corresponding impact coefficient of the drug, combined with the cumulative effect intensity of the drug and the preset impact weight, so as to predict the risk probability of each side effect. Based on the cumulative effect intensity of the drug, the preset influence weights of the drug on clinical indicators, and the differences between the time-series data of each clinical indicator after drug administration and the preset comparison time-series data, an elastic network regression model is established to determine the influence coefficient of the drug on each clinical indicator. The influence coefficient includes the main influence coefficient and the interaction influence coefficient, including: Analyze the differences between the time-series data of each clinical indicator after medication and the preset comparison time-series data to determine the indicator offset of each clinical indicator; Combine any two drugs in pairs to obtain all unique drug combinations; A regression function is constructed based on the cumulative effect intensity of drugs, the cumulative effect intensity of drugs in drug combinations, the preset influence weight of drugs on each clinical indicator, the preset main influence coefficient of each drug, and the preset interaction influence coefficient of drug combinations. The preset main influence coefficient and the preset interaction influence coefficient are variable, and the preset main influence coefficient is not 0. In the elastic network regression model, the value of the regression function with the smallest absolute value of the difference between the index offset and each clinical indicator is taken as the objective function value. The preset main influence coefficient and preset interaction influence coefficient of each drug under the objective function value are used as the main influence coefficient of each drug and the interaction influence coefficient of the drug combination to which it belongs.

2. The drug side effect risk prediction system for tumor patients based on multi-source data according to claim 1, characterized in that, Based on the principle of first-order elimination kinetics, the attenuation factor of each drug is obtained according to its elimination rate constant and duration of action, including: The absolute value of the time difference between two consecutive medication times is taken as the duration of drug action at the time of the previous medication. The attenuation factor of each drug is obtained by taking the negative of the product of the duration of drug action at each administration and the elimination rate constant of each drug, and using the natural constant as the base.

3. The drug side effect risk prediction system for tumor patients based on multi-source data according to claim 1, characterized in that, The method for obtaining the cumulative effect intensity includes: After each administration, the ratio of the dose of each drug to the median lethal dose was used as the normalized dose, and the product of the normalized dose of each drug and the attenuation factor was used as the cumulative effect factor. The normalized value of the cumulative effect factor corresponding to each drug after all administrations is taken as the cumulative effect intensity of each drug.

4. The drug side effect risk prediction system for tumor patients based on multi-source data according to claim 1, characterized in that, The method for obtaining the indicator offset includes: For each clinical indicator, the difference between the mean value of the medication time series data after each medication administration and the mean value of the preset comparison time series data is calculated, which is used as the indicator shift factor for each clinical indicator after each medication administration. The mean of the index offset factor for each clinical indicator across all drug administration times is used as the index offset for each clinical indicator.

5. The drug side effect risk prediction system for tumor patients based on multi-source data according to claim 1, characterized in that, The formula model of the regression function includes: in, This represents the value of the j-th clinical indicator under the regression function; This represents the preset influence weight of the i-th drug on the j-th clinical indicator; This represents the pre-defined main influence coefficient of the i-th drug on the j-th clinical indicator; This represents the cumulative effect intensity of the i-th drug; n represents the number of drug types; m represents the total number of drug combinations. This represents the interaction coefficient between the two drugs in the k-th drug combination; This represents the cumulative effect intensity of drug 1 in the k-th drug combination; This represents the cumulative effect intensity of drug 2 in the k-th drug combination; This is the random error term.

6. The drug side effect risk prediction system for tumor patients based on multi-source data according to claim 1, characterized in that, The process of updating the decision tree's decision order by utilizing the drug's corresponding influence coefficient, combined with the drug's cumulative effect intensity and preset influence weights, to predict the risk probability of each side effect includes: The abnormal contribution of each drug to each clinical indicator is determined based on the main influence coefficient of each drug on each clinical indicator and the interaction influence coefficient of the drug combination to which it belongs. Determine the corresponding clinical indicators for each type of adverse reaction; Based on the random forest algorithm, for any side reaction, several decision trees are constructed. In each decision tree, the decision order of the decision tree is updated by the descending order of the abnormal contribution, thereby outputting the risk factor of that side reaction. The risk probability of this type of side reaction is obtained by normalizing the mean of the risk factors under all decision trees.

7. The drug side effect risk prediction system for tumor patients based on multi-source data according to claim 6, characterized in that, The method for obtaining the abnormal contribution includes: Choose one drug as the test drug and choose one clinical indicator as the test indicator. Under the target indicator, the product of the absolute value of the main influence coefficient of the test drug, the cumulative effect intensity of the test drug, and the preset influence weight of the test drug on the target indicator is used as the first abnormal contribution factor of the test drug to the target indicator. Under the test index, when the interaction coefficient of all drug combinations to which the test drug belongs is 0, the normalized value of the first abnormal contribution factor of the test drug to the test index is taken as the abnormal contribution degree of the test drug to the test index. Under the target indicator, when the interaction coefficients of all drug combinations to which the target drug belongs are not all zero, the drug combinations with non-zero interaction coefficients are taken as the target combinations. In each target combination, the proportion of the cumulative effect intensity of the target drug is analyzed. Then, combined with the interaction coefficients and the preset influence weights of the target drug on the target indicator, the second abnormal contribution factor of the target drug on the target indicator is obtained. The normalized value of the sum of the first and second abnormal contribution factors of the target drug on the target indicator is taken as the abnormal contribution degree of the target drug on the target indicator.

8. The drug side effect risk prediction system for tumor patients based on multi-source data according to claim 7, characterized in that, The method for obtaining the second abnormal contribution factor includes: Under the test index, in each test combination, the proportion of the cumulative effect intensity of the test drug is used as the scaling factor, and the product of the scaling factor and the absolute value of the interaction coefficient corresponding to the test combination is used as the adjustment factor. The sum of the adjustment factors of all test combinations and the product of the preset influence weight of the test drug on the test indicator are used as the second abnormal contribution factor of the test drug to the test indicator.

9. The drug side effect risk prediction system for tumor patients based on multi-source data according to claim 1, characterized in that, The preset comparison time series data consists of time series data on the changes of various clinical indicators under the condition of no medication.

Images