JP7433503B1 - High-throughput real-world drug efficacy and safety evaluation methods and systems - Google Patents

Abstract

The present invention provides a method and system for evaluating drug efficacy and safety in a high-throughput real-world setting. [Solution] The method includes the steps of constructing a dataset, constructing a drug efficacy/safety indicator phenotype library, selecting a target drug, and identifying a target event, an outcome event, a target date, and an outcome date. the step of defining and extracting patient ID, target date and outcome date, selecting indicators from the drug efficacy/safety indicator phenotype library, obtaining target drug-efficacy/safety indicator pairs, and creating an event sequence. Perform high-throughput signal screening of the target drug-efficacy/safety indicator pair using symmetry analysis to obtain a primary screening positive signal, and perform causal evaluation on the primary screening positive signal to determine the effectiveness and efficacy of the target drug. and determining safety. [Selection diagram] Figure 1

Description

The present invention relates to the technical field of drug management, and particularly to a high-throughput real-world evaluation method and system for drug efficacy and safety.

Drug evaluation is an important tool to promote rational dosing in clinical practice and effectively control pharmaceutical costs. Medical research is focused on using the highest quality evidence to determine the therapeutic efficacy and safety of new therapeutic substances and maximizing the time it takes to bring drugs to market. There is. The use of randomized controlled trials (RCTs) revolutionized the development of medicine in the 20th century, providing a gold standard for evaluating the therapeutic efficacy of drugs. However, randomized controlled trials are not always feasible, and recruiting subjects is extremely difficult and time-consuming, especially in severe or rare diseases. The amount and availability of routine medical data collected electronically during the actual diagnosis and treatment process (also referred to as “real-world data (RWD)”) has increased rapidly in recent years. , it can be used as a new method to assess the safety and efficacy of drugs after release. Compared to clinical trials, RWD often reflects the actual situation of treatment as it includes more patients with different characteristics, and therefore studies using RWD reflect the actual population that needs to be treated in clinical practice. can be better represented. Drug evaluation by RWD helps to obtain evidence of the actual clinical application situation after drug release, promoting the safety and effectiveness of patient medication, while also supporting the research and development of new drugs and new clinical trials of new uses of old drugs. can provide enlightenment for the design of

In recent years, countries around the world have issued policies and laws to promote the research, development, and launch of new drugs using RWD. There is no framework yet. In terms of effectiveness, the system described in the invention patent ``System for evaluating drug treatment effects based on real-world research'' with publication number CN114025253A has the ability to collect real-world medical data by realizing wide-area VNA barrier-free data collection. The present invention focuses on the system network configuration of a drug evaluation system and discloses how to use real-world data to evaluate therapeutic effects. Not yet. Regarding the safety aspect, the invention patent with the registration number CN111402971B "Method and system for rapid identification of adverse drug reactions based on big data" and the invention patent with the registration number CN104765947B "Potential drug adverse reactions towards big data" Both reaction data mining methods discover or evaluate adverse drug reactions based on reported adverse drug reaction information, combining field knowledge and the degree of association between drugs and adverse reactions in real-world data. The invention patent "Method and device for identifying adverse drug reactions" with application number CN202111628626.3 identifies adverse reactions by a method called prescription sequence symmetry analysis, and its basic principle is to use labeled drugs as surrogate events for adverse drug reactions. and to identify adverse drug reactions by analyzing before and after use of the index drug and label drug.

The above content has the following limitations. 1) Drug-adverse reaction related signals are from the drug adverse reaction reporting system, and many signals that actually occurred in the real world but were not noticed are missing; 2) signals obtained by combining domain knowledge and real-world data 3) Prescription sequence symmetry analysis is always applied to insurance databases and surrogate drug selection. seriously impacts the reliability of algorithm outcomes.

Therefore, we propose a high-throughput real-world evaluation method and system for drug efficacy and safety.

In order to solve the above technical problems, the present invention provides a high-throughput real-world evaluation method and system for drug efficacy and safety.

The technical solution used by the present invention is as follows.

A high-throughput real-world evaluation method for drug efficacy and safety, the method comprising:
Step S1: acquiring real world data, performing integrated data coding on the real world data, extracting the minimum necessary data of the real world data after the integrated data coding, and completing the construction of a dataset;
Step S2 of constructing a triplet mode including a concept set, a behavior, and a standard, and constructing a drug efficacy/safety indicator phenotype library using the triplet mode, a drug efficacy/safety term set, and a clinical index; ,
Select a target drug, select an indicator from the efficacy/safety indicator phenotype library of the drug, obtain a target drug-efficacy/safety indicator pair, and specify the target event, outcome event, target date, and outcome date. step S3 of defining a data set, extracting patient ID, target date, and outcome date, and configuring a data set for performing high-throughput signal primary screening;
Performing high-throughput signal screening of the target drug-efficacy/safety indicator pair on the data set of the high-throughput signal primary screening using event sequence symmetry analysis to obtain a positive signal for the primary screening, step S4;
The step S5 includes performing a causal evaluation on the positive signal of the primary screening to determine the effectiveness and safety of the target drug.

Furthermore, the step S3 specifically includes:
The target event is the first use of the target drug, one or more indicators selected by the user from the drug efficacy/safety indicator phenotype library are configured as outcome indicators, and the first occurrence of each of the outcome indicators is configured as an outcome indicator. a substep S31 in which the outcome event is set as an outcome event;
a substep S32 in which the date of occurrence of the target event is set as a target date, and the date of occurrence of the outcome event is set as an outcome date;
Select the target event and the outcome event, screen patients in the dataset in which both the target event and the outcome event occur, extract the patient ID, target date, and outcome date, and perform high-throughput signal primary screening. and a sub-step S33 of configuring a data set for performing.

Furthermore, the step S4 specifically includes:
a sub-step S41 of calculating an ordinal ratio using a ratio between the number of people for whom the outcome event occurred after the target event occurred and the number of people for whom the outcome event occurred before the target event occurred; ,
a sub-step S42 of calculating an ineffective ordinal ratio and adjusting the ordinal ratio using the ineffective ordinal ratio to obtain a modified ordinal ratio;
a sub-step S43 of calculating a confidence interval of the modified ordinal ratio by calculating a confidence interval of the probability that the outcome event occurs after the target event occurs;
and a substep S44 in which a signal in which the lower limit of the confidence interval of the corrected ordinal ratio is greater than 1 is determined as a primary screening positive signal.

Furthermore, the sub-step S42 specifically includes:
a sub-step S421 of calculating an overall average probability of the outcome event occurring within a preset observation window after the target event occurs for the data set performing high-throughput signal primary screening;
a sub-step S422 of calculating a no-effect order ratio using the overall average probability;
and a substep S423 of calculating a ratio between the ordinal ratio and the no-effect ordinal ratio to obtain a modified ordinal ratio.

Furthermore, the step S5 specifically includes:
Screening patients in the data set in which the event of interest occurred against the target drug-efficacy/safety indicator pair corresponding to the primary screening positive signal, and placing the patient in a user queue and an initial non-user queue. a substep S51 of allocating;
a sub-step S52 of randomly selecting one type of drug as an alternative drug from among the drugs used by the patient in the initial non-user cue to construct a non-user cue;
The first use of the target drug or the alternative drug is the registration time, the use of the target drug or the alternative drug is the treatment method, and whether or not the outcome event occurs is the result, before the patient registers. extracting baseline data of the data set, configuring the demographic information of the data set and the baseline data as a covariable set, extracting features, screening variables used to calculate the average treatment effect. A substep S53 of obtaining
a sub-step S54 of calculating the average therapeutic effect of the target drug using the variables used to calculate the average therapeutic effect;
Repeat substeps S52 to S54 until the maximum number of tests is reached, calculate the average effect of all the average therapeutic effects, and calculate the confidence interval of the average effect to evaluate the effectiveness and safety of the target drug. and a sub-step S55 for determining.

Furthermore, the sub-step S51 specifically includes:
Screening the patient in the data set in which the target event occurred for the target drug-efficacy/safety indicator pair corresponding to the primary screening positive signal, and obtaining the main diagnosis received by the patient; a sub-step S511 that statistics the patients based on ICD-10 codes and sets the main diagnosis with the largest number of patients as the basic diagnosis;
and assigning the patient to a user queue and an initial non-user queue.

Furthermore, the sub-step S53 specifically includes:
a sub-step S531 of selecting one covariate from the set of covariates, marking the covariate as an importance covariate, and marking the remaining variables of the set of covariables as remaining variables;
An outcome event prediction model using the overall covariate as an input variable is constructed for the conditional expected value of the result under conditions given the importance covariate and the remaining variables, and the condition of the result is a substep S532 of obtaining an estimated value of expected value;
a substep S533 of estimating the probability that the value of the importance covariate is 1 under conditions given the remaining variables and obtaining an estimated value;
a sub-step S534 of constructing a variable importance indicating disturbance variable using the estimated value of the probability that the importance covariate is 1;
a substep S535 of constructing an outcome event regression model using the estimated value of the conditional expected value of the result as an intercept term and the variable importance indicating disturbance variable as an input variable, and obtaining a variable importance disturbance parameter;
a sub-step S536 of updating the estimated value of the conditional expected value of the result using the variable importance disturbance parameter and the variable importance indicating disturbance variable, and calculating variable importance;
a sub-step S537 of calculating the standard deviation and confidence interval of the variable importance;
Substeps S531 to S537 are repeated until each covariate in the covariate set is traversed, and the average treatment effect is calculated for variables whose left and right confidence intervals of the standard deviation of variable importance do not include zero. and a substep S538 in which variables are used.

Furthermore, the sub-step S54 specifically includes:
a substep S541 of constructing an outcome event prediction model using the variables used to calculate the treatment method and average treatment effect as input variables, and obtaining an estimated value of the conditional expected value of the outcome event;
estimating the probability of treatment assignment to obtain an estimate of treatment assignment S542;
a sub-step S543 of constructing an instruction disturbance variable using the estimated value of the treatment allocation;
A regression model of the outcome event and the indicated disturbance variable is constructed using the estimated value of the conditional expected value of the outcome event calculated by inputting the variables of the treatment method and the average treatment effect as an intercept term, and the estimated disturbance parameter is A substep S544 of obtaining
The method includes a substep S545 of updating the estimated value of the conditional expected value of the outcome event using the estimated disturbance parameter and the indicated disturbance variable, and calculating an average treatment effect.

Furthermore, the sub-step S55 specifically includes:
a substep S551 of repeating substeps S52 to S54 until the maximum number of tests is reached, and calculating the average effect of all the average treatment effects;
Calculating the confidence interval of the average effect, obtaining a signal with a left confidence interval larger than zero or a signal with a right confidence interval smaller than zero as a final positive signal, and determining the efficacy and safety of the target drug. The substep S552 includes the substep S552.

The present invention further provides a high-throughput real-world drug efficacy and safety evaluation system,
a data collection module for retrieving data from existing data and cleaning said data to complete the construction of the dataset;
a high-throughput signal screening module for converting data in the dataset into clinical events and performing high-throughput signal screening with event sequence symmetry analysis to obtain a primary screening positive signal;
a causal evaluation module for performing causal evaluation on the positive signal of the primary screening and determining the effectiveness and safety of the target drug;
and a result display module for displaying efficacy and safety results of the target drug.

The beneficial effects of the present invention are as follows.

1. The present invention is based on a large-scale real-world data set, is not limited to a specific drug and specific efficacy/safety indicators, and uses a causal assessment algorithm based on high-throughput signal screening and clinical trial simulation to evaluate the target drug. Causal evaluation of drug efficacy can be completed in a high-throughput manner.

2. The present invention constructs a scalable drug efficacy/safety phenotype library based on the triplet mode of "concept set + action + standard", which is used for high-throughput real-world drug evaluation.

3. In evaluating the causal relationship between the target drug and the outcome event, the present invention uses two-stage target maximum likelihood estimation. First, the importance of all data features generated in the real world of patient cues is evaluated, and human For data-driven and not dependent on expert knowledge, select the covariates most relevant to the outcome event and treatment assignment, then calculate and restore the average treatment effect based on a comparison of these variables. By randomly selecting an alternative drug and simulating a clinical trial multiple times, we can obtain reliable average treatment effect results, thereby establishing a quantitative causal relationship between the target drug and the outcome measure. Get a grade.

1 is a flowchart of a high-throughput real-world drug efficacy and safety evaluation method according to the present invention. FIG. 1 is a schematic diagram of a high-throughput real-world drug efficacy and safety evaluation system according to the present invention.

The following description of at least one exemplary embodiment is only illustrative in nature and in no way limits the invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without the need for creative efforts fall within the protection scope of the present invention.

As shown in FIG. 1, the high-throughput real-world evaluation method for drug efficacy and safety includes steps S1 to S5.

In step S1, real world data is acquired, integrated data coding is performed on the real world data, and after the integrated data coding, the minimum necessary data of the real world data is extracted to complete the construction of the dataset. .

In step S2, a triplet mode of concept set + action + standard is constructed, and a drug efficacy/safety indicator phenotype library is constructed using the triplet mode, drug efficacy/safety term set, and clinical index.

In step S3, a target drug is selected, an index is selected from the efficacy/safety index phenotype library of the drug, a target drug-efficacy/safety index pair is obtained, and a target event, an outcome event, and a target date are obtained. , define the outcome date, extract the patient ID, target date, and outcome date, and construct a dataset for high-throughput signal primary screening.

In step S31, the first use of the target drug is set as a target event, one or more indicators selected by the user from the drug efficacy/safety indicator phenotype library are configured as outcome indicators, and each of the outcome indicators The first occurrence of is the outcome event.

In step S32, the date of occurrence of the target event is set as the target date, and the date of occurrence of the outcome event is set as the outcome date.

In step S33, the target event and the outcome event are selected, patients in the data set in which both the target event and the outcome event occur are screened, patient IDs, target dates, and outcome dates are extracted, and high-throughput Construct a data set for primary signal screening.

In step S4, high-throughput signal screening of the target drug-efficacy/safety indicator pair is performed on the data set of the high-throughput signal primary screening using event sequence symmetry analysis to obtain a positive signal for the primary screening. .

In step S41, an ordinal ratio is calculated using the ratio of the number of people for whom the outcome event occurred after the target event occurred and the number of people for whom the outcome event occurred before the target event occurred. .

In step S42, an ineffective ordinal ratio is calculated, and the ordinal ratio is adjusted using the ineffective ordinal ratio to obtain a corrected ordinal ratio.

In step S421, the overall average probability that the outcome event occurs within a preset observation window after the target event occurs is calculated for the data set that performs high-throughput signal primary screening.

In step S422, the no-effect order ratio is calculated using the overall average probability.

In step S423, a ratio between the order ratio and the no-effect order ratio is calculated to obtain a corrected order ratio.

In step S43, a confidence interval for the modified ordinal ratio is calculated by calculating a confidence interval for the probability that the outcome event will occur after the target event occurs.

In step S44, a signal for which the lower limit of the confidence interval of the corrected ordinal ratio is greater than 1 is determined to be a positive signal for the primary screening.

In step S5, a causal evaluation is performed on the positive signal of the primary screening to determine the effectiveness and safety of the target drug.

In step S51, a patient in the data set in which the target event has occurred is screened for the target drug-efficacy/safety index pair corresponding to the primary screening positive signal, and the patient is screened for the user cue and the initial Assign to non-user queues.

In step S511, the patient in which the target event occurred in the data set is screened for the target drug-efficacy/safety index pair corresponding to the primary screening positive signal, and the main diagnosis that the patient received is screened. is obtained, statistics are made on the patients based on the ICD-10 code, and the main diagnosis with the largest number of patients is set as the basic diagnosis.

In step S512, the patient is assigned to a user queue and an initial non-user queue.

In step S52, one type of drug is randomly selected as an alternative drug from among the drugs used by the patient in the initial non-user queue, and a non-user queue is constructed.

In step S53, the first use time of the target drug or the alternative drug is set as a registration time, the use of the target drug or the alternative drug is set as a treatment method, and whether or not the outcome event occurs is set as a result, and the patient extracting baseline data before enrollment, configuring the demographic information of the dataset and the baseline data as a covariate set, extracting features, screening and calculating the average treatment effect. Get the variables used.

In step S531, one covariate is selected from the covariable set, the covariate is designated as an importance covariate, and the remaining variables in the covariable set are designated as remaining variables.

In step S532, under conditions given the importance covariate and the remaining variables, construct an outcome event prediction model using the overall covariate as an input variable for the conditional expected value of the result; Obtaining an estimate of the conditional expectation of the result.

In step S533, the probability that the value of the importance covariate is 1 is estimated under the conditions given the remaining variables, and an estimated value is obtained.

In step S534, a variable importance indicating disturbance variable is constructed using the estimated value of the probability that the importance covariate is 1.

In step S535, the estimated value of the conditional expected value of the result is used as an intercept term, the variable importance indicating disturbance variable is used as an input variable, an outcome event regression model is constructed, and a variable importance disturbance parameter is obtained.

In step S536, the estimated value of the conditional expected value of the result is updated using the variable importance disturbance parameter and the variable importance indicating disturbance variable, and the variable importance is calculated.

In step S537, the standard deviation and confidence interval of the variable importance are calculated.

In step S538, steps S531 to S537 are repeated until each covariate in the covariate set is traversed, and the average treatment effect is calculated for variables whose left and right confidence intervals of the standard deviation of variable importance do not include zero. Let it be a variable used in

In step S54, the average therapeutic effect of the target drug is calculated using the variables used to calculate the average therapeutic effect.

In step S541, an outcome event prediction model is constructed using the treatment method and the variables used to calculate the average treatment effect as input variables, and an estimated value of the conditional expected value of the outcome event is obtained.

In step S542, the probability of treatment assignment is evaluated to obtain an estimate of treatment assignment.

In step S543, an instruction disturbance variable is constructed using the estimated value of the treatment assignment.

In step S544, a regression model of the outcome event and the indicated disturbance variable is constructed using the estimated value of the conditional expected value of the outcome event calculated by inputting the variables of the treatment method and the average treatment effect as an intercept term, Obtain estimated disturbance parameters.

In step S545, the estimated value of the conditional expected value of the outcome event is updated using the estimated disturbance parameter and the indicated disturbance variable, and the average treatment effect is calculated.

In step S55, steps S52 to S54 are repeated until the maximum number of tests is reached, the average effect of all the average therapeutic effects is calculated, and the confidence interval of the average effect is calculated to determine the effectiveness of the target drug. Determine safety.

In step S551, steps S52 to S54 are repeated until the maximum number of tests is reached, and the average effect of all the average treatment effects is calculated.

In step S552, a confidence interval of the average effect is calculated, and a signal with a left confidence interval larger than zero or a signal with a right confidence interval smaller than zero is obtained as a final positive signal, and the effectiveness of the target drug is determined. Determine safety.

As shown in Figure 2, the high-throughput real-world drug efficacy and safety evaluation system
a data collection module for retrieving data from existing data and cleaning said data to complete the construction of the dataset;
a high-throughput signal screening module for converting data in the dataset into clinical events and performing high-throughput signal screening with event sequence symmetry analysis to obtain a primary screening positive signal;
a causal evaluation module for performing causal evaluation on the positive signal of the primary screening and determining the effectiveness and safety of the target drug;
and a result display module for displaying efficacy and safety results of the target drug.

In an embodiment, a high-throughput real-world evaluation method for drug efficacy and safety includes steps S1 to S5.

In step S1, real world data is acquired, integrated data coding is performed on the real world data, and after the integrated data coding, the minimum necessary data of the real world data is extracted to complete the construction of the dataset. .

Acquire real world data, acquire medication data, diagnostic data, surgical data, laboratory test results, etc. from electronic medical record data, do not process the time of data occurrence, retain the original date and time, specifically, The information obtained includes (1) demographic information including date of birth, gender, and ethnicity, (2) basic medical information including allergy history, family history, and blood type, (3) diagnostic records, chemical test results, Contains diagnostic and treatment information, including medication records, surgical records, and imaging reports.

First, integrated data coding was performed, and the self-defined codes of this example were used for gender, age, ethnicity, blood type, chemical test items, and medication information, and ICD-10 codes were used for diagnosis and family disease history. ICD-9-CM codes are used for surgical information.

The data set described in step S1 includes demographic information, chemical test results, medication information, diagnostic information, surgical records, and clinical symptoms.

After integrated data coding, first extract the minimum necessary data from the structured data and use it for clinical event conversion. The minimum required structured data includes the following:
Demographic information: patient ID, date of birth, ethnicity, blood type,
Chemical test results: patient ID, chemical test item code, chemical test event, chemical test result,
Medication information: patient ID, drug code, medication start time,
Diagnosis information: patient ID, diagnosis code, diagnosis time,
Surgery record: patient ID, surgery code, surgery time,
First, body parts and clinical symptoms are extracted from the natural language text information in the image test report using named entity extraction technology, and after mapping of self-configured codes is completed, the information is converted to the minimum necessary data.
Clinical symptoms: patient ID, clinical symptom code, clinical symptom time.

In step S2, a triplet mode of concept set + action + standard is constructed, and a drug efficacy/safety indicator phenotype library is constructed using the triplet mode, drug efficacy/safety term set, and clinical index.

The efficacy/safety index phenotype of a drug is different for different drugs and different evaluation dimensions, therefore, in this example, a triplet mode of "concept set + action + standard" is proposed. A "concept set" consists of one or more data codes (e.g., diagnostic code, chemical test code, etc.), and an "action" consists of a series of actions or conditions, such as "occurrence,""morethan,"""Lessthan","Equal","Increase","Decrease", and "Standard" refers to the level of a specific efficacy/safety indicator and may be omitted. below,
"Tumor size" + "reduction" + "0.5 cm",
Two examples are “serum alanine aminotransferase” + “more than” + “3 times the upper limit of normal value”.
The above triplet mode can be used to build a drug efficacy/safety phenotype library based on existing drug efficacy/safety term sets and clinically accepted indicators in clinical guidelines. It is also possible to predefine personalization for specific efficacy/safety indicators according to the user's needs to improve the drug efficacy/safety indicator phenotype library.

In step S3, a target drug is selected, an index is selected from the efficacy/safety index phenotype library of the drug, a target drug-efficacy/safety index pair is obtained, and a target event, an outcome event, and a target date are obtained. , define the outcome date, extract the patient ID, target date, and outcome date, and construct a dataset for high-throughput signal primary screening.

In step S31, the first use of the target drug is set as a target event, one or more indicators selected by the user from the drug efficacy/safety indicator phenotype library are configured as outcome indicators, and each of the outcome indicators The first occurrence of is the outcome event.

The target drug may be a single drug or a class of drugs with the same therapeutic effect or the same properties, and when a class of drugs is selected as the target drug, the selected drugs may be the same drug. considered to be.

In step S32, the date of occurrence of the target event is set as the target date, and the date of occurrence of the outcome event is set as the outcome date.

In step S33, the target event and the outcome event are selected, patients in the data set in which both the target event and the outcome event occur are screened, patient IDs, target dates, and outcome dates are extracted, and high-throughput Construct a data set for primary signal screening.

In step S4, high-throughput signal screening of the target drug-efficacy/safety indicator pair is performed on the data set of the high-throughput signal primary screening using event sequence symmetry analysis to obtain a positive signal for the primary screening. .

In step S41, the order ratio sr is calculated using the ratio of the number of people for whom the outcome event occurred after the target event occurred and the number of people for whom the outcome event occurred before the target event occurred. do.

Event sequence symmetry analysis can be affected by changing trends in medical practice over time, for example, changes in health insurance policies that cause a sudden increase or decrease in certain chemical tests or medications within a certain time period. Both of these can lead to erroneous effect estimates. To correct for such deviations, a no-effect order ratio null_effect_sr can be calculated, which adjusts the order ratio sr based on the order probability that the outcome event occurs after the target event in the absence of any causal relationship. do.

In step S42, an ineffective ordinal ratio is calculated, and the ordinal ratio is adjusted using the ineffective ordinal ratio to obtain a corrected ordinal ratio.

In step S421, the overall average probability p _a of the outcome event occurring within a preset observation window after the target event occurs is calculated for the data set on which high-throughput signal primary screening is performed. .

The calculation of the no-effect ordinal ratio null_effect_sr is derived from the probability p that the outcome event occurs within a specified observation window after the event of interest occurs, for each patient in which the event of interest occurs.

where d is the time length of the continuous observation window, x is the day on which the event of interest occurred, and O _t is the number of patients in which the outcome event occurred on day t.

For the entire patient group, the overall average probability p _a of the outcome event occurring after the target event is calculated by weighting the number of patients in which the target event occurred for m consecutive days within the observation period, and then The purpose is to find the average value for the days.

Here, u indicates the last day of the observation period, O _t indicates the number of people for whom the outcome event occurred on that day, T _m indicates the number of people for whom the target event occurred on day m, and d indicates the number of people for whom the outcome event occurred on day m. is the time length of a continuous observation window.

In step S422, the no-effect order ratio null_effect_sr is calculated using the overall average probability.

In step S423, a ratio between the ordinal ratio and the ineffective ordinal ratio is calculated to obtain a modified ordinal ratio adjusted_sr.

の信頼区間を計算することにより、前記修正された順序比ａｄｊｕｓｔｅｄ＿ｓｒの信頼区間を計算する。 In step S43, the probability that the outcome event will occur after the target event occurs is determined.

A confidence interval for the modified ordinal ratio adjusted_sr is computed by computing a confidence interval for the modified order ratio adjusted_sr.

を対象イベントが発生した後にアウトカムイベントが発生する確率とし、すなわち、

であると、

が得られる。

Let be the probability that the outcome event occurs after the target event occurs, i.e.,

So,

is obtained.

の９５％信頼区間を計算することにより修正された順序比ａｄｊｕｓｔｅｄ＿ｓｒの信頼区間を計算する。本実施例では、対象イベントが発生した後にアウトカムイベントが発生する確率

の信頼区間の計算は下式を使用する。

Probability that the outcome event will occur after the target event occurs

Calculate the confidence interval of the modified ordinal ratio adjusted_sr by calculating the 95% confidence interval of adjusted_sr. In this example, the probability that the outcome event will occur after the target event occurs is

To calculate the confidence interval of , use the following formula.

は１．９６を取る。 where n is the total number of patients participating in the calculation,

takes 1.96.

In step S44, a signal for which the lower limit of the confidence interval of the corrected ordinal ratio is greater than 1 is determined to be a positive signal for the primary screening.

In step S5, a causal evaluation is performed on the positive signal of the primary screening to determine the effectiveness and safety of the target drug.

In step S51, a patient in the data set in which the target event has occurred is screened for the target drug-efficacy/safety index pair corresponding to the primary screening positive signal, and the patient is screened for the user cue and the initial Assign to non-user queues.

In step S511, the patient in which the target event occurred in the data set is screened for the target drug-efficacy/safety index pair corresponding to the primary screening positive signal, and the main diagnosis that the patient received is screened. is obtained, statistics are made on the patients based on the ICD-10 code, and the main diagnosis with the largest number of patients is designated as the basic diagnosis and marked as D.

In step S512, the patient is assigned to a user queue and an initial non-user queue.

Screen patients whose primary diagnosis is the underlying diagnosis from the data set and assign them to a user queue if they meet the following inclusion criteria: (1) the patient has been taking the drug continuously (e.g., less than 30 days have passed between two doses), and (2) the time the patient first obtained the prescription must be in the baseline diagnostic record. (3) the patient has at least one year (365 days) of diagnostic and treatment records in the dataset before the first prescription is given. Patients whose main diagnosis is the underlying diagnosis are screened from the data set and assigned to the initial non-user queue if they are not using the target drug.

In step S52, one type of drug is randomly selected as an alternative drug from among the drugs used by the patient in the initial non-user queue, and a non-user queue is constructed.

When trying to estimate the effect of a drug, it is necessary to compare user cues and non-user cues to which alternative drugs are assigned. Once an alternative drug is identified, non-user cues are defined by the same built-in standards as above, but related to the alternative drug. To avoid overlap between user and non-user queues, the present embodiment further excludes any patients using the target drug from the non-user queue. In this example, the alternative drug is randomly selected from all drugs used by patients in the non-user queue and does not include the target drug itself. Such non-user cues directly compare the drug of interest to drugs with the same therapeutic indication, thereby reducing confounding by different indications.

In step S53, the first use time of the target drug or the alternative drug is set as a registration time, the use of the target drug or the alternative drug is set as a treatment method, and whether or not the outcome event occurs is set as a result, and the patient extracting baseline data before enrollment, configuring the demographic information of the dataset and the baseline data as a covariate set, extracting features, screening and calculating the average treatment effect. Get the variables used.

Diagnosis data is extracted from within one year before registration time, chemical test results, surgery, and clinical symptoms are extracted from within one month before registration time, and medication information is data from within 7 days before registration time. Extract. A first stage of targeted maximum likelihood estimation is then activated to assess the importance of each baseline variable and further screen for confounding factors used in assessing the average treatment effect.

In step S531, one covariate is selected from the covariable set, the covariate is designated as an importance covariate, and the remaining variables in the covariable set are designated as remaining variables.

が重要性を評定しようとする重要性共変数であり、

が残りの変数であるとする。 In the dataset, the data structure is O = (W, A, Y), where W = W ₁ , W ₂ , ..., W _k is a covariate set, and in this example, demographic information and extracted A is the treatment method, i.e., whether to use the target drug or an alternative drug, and Y is the outcome, i.e., whether the outcome event occurs or not. .

is the importance covariate whose importance is to be rated, and

Let be the remaining variables.

In step S532, an outcome event prediction model using the overall covariate as an input variable is constructed for the conditional expected value E of the result under conditions given the importance covariate and the remaining variables. , obtain an estimate of the conditional expectation of the result.

、残りの変数

が与えられた条件の下で、結果Ｙの条件付き期待値

に対してフィッティング推定を行い、すなわち全体共変数Ｗを入力変数とするアウトカムイベント予測モデル

を構築し、重要性を評定しようとする各重要性共変数

に対して、

の値は結果Ｙの条件付き期待値

の推定値であり、ロジスティック回帰等の任意の機械学習分類方法を使用して、

である場合の結果Ｙの条件付き期待値の推定値

及び

である場合の結果Ｙの条件付き期待値の推定値

をそれぞれ計算することができる。 importance covariate

, remaining variables

The conditional expectation value of the outcome Y under the given conditions

In other words, the outcome event prediction model uses the overall covariate W as an input variable.

and evaluate the importance of each covariate.

For,

The value of is the conditional expected value of the outcome Y

is an estimate of , using any machine learning classification method such as logistic regression,

Estimate of the conditional expectation of outcome Y if

as well as

Estimate of the conditional expectation of outcome Y if

can be calculated respectively.

In step S533, the probability that the value of the importance covariate is 1 is estimated under the conditions given the remaining variables, and an estimated value is obtained.

が設定された前提下で、重要性共変数

が１である確率

を推定し、

と記し、ロジスティック回帰等の任意の機器学習分類方法を使用して、モデルをフィッティングした後に、残りの変数

が設定された前提下で、重要性共変数

が１である確率の推定値

及び残りの変数

が設定された前提下で、重要性共変数

が０である確率の推定値

を算出することもできる。 remaining variables

The importance covariate under the assumption that

The probability that is 1

Estimate

and after fitting the model using any instrumental learning classification method such as logistic regression, the remaining variables

The importance covariate under the assumption that

estimate of the probability that is 1

and the remaining variables

The importance covariate under the assumption that

estimate of the probability that is 0

It is also possible to calculate

In step S534, a variable importance indicating disturbance variable is constructed using the estimated value of the probability that the importance covariate is 1.

により変数重要性指示外乱変数Ｈ_ｊを構築する。

A variable importance indicating disturbance variable H _j is constructed by .

である場合、

である。 Here, I is an indicator function,

If it is,

It is.

In step S535, the estimated value of the conditional expected value of the result is used as an intercept term, the variable importance indicating disturbance variable is used as an input variable, an outcome event regression model is constructed, and a variable importance disturbance parameter is obtained.

と結果Ｙの条件付き期待値

との間に偏差があり、変数重要性外乱パラメータε_ｊで

を修正する必要がある。使用方法については、

の値を切片項とし、アウトカムイベントと変数重要性指示外乱変数の回帰モデル、すなわち

を構築し、モデルをフィッティングした後に、変数重要性指示外乱変数

の係数は変数重要性外乱パラメータε_ｊである。 At this time, in this embodiment,

and the conditional expected value of the outcome Y

There is a deviation between the variable importance disturbance parameter ε _j

needs to be corrected. For instructions on how to use,

A regression model of the outcome event and the disturbance variable indicating variable importance, with the value of as the intercept term, i.e.

After building and fitting the model, the variable importance indicates the disturbance variable

The coefficient of is the variable importance disturbance parameter ε _j .

In step S536, the estimated value of the conditional expected value of the result is updated using the variable importance disturbance parameter and the variable importance indicating disturbance variable, and the variable importance is calculated.

の更新を完了する。

Complete the update.

Variable importance is calculated using the following formula.

where n is the number of patients in the dataset and i is the i-th record in the dataset.

In step S537, the standard deviation and confidence interval of the variable importance are calculated.

In order to complete the variable screening, it is necessary to calculate the standard deviation of variable importance, and the calculation formula is as follows:

であり、Ｖａｒは標本分散を示し、従って、ＶＩ_ｊの９５％信頼区間は

である。 here,

, Var indicates the sample variance, and therefore the 95% confidence interval of VI _j is

It is.

In step S538, steps S531 to S537 are repeated until each covariate in the covariate set is traversed, and the average treatment effect is calculated for variables whose left and right confidence intervals of the standard deviation of variable importance do not include zero. Let the variable W _important be used for

の変数を、更に平均治療効果を行うための変数Ｗ_{ｉｍｐｏｒｔａｎｔ}とする。 left and right confidence interval

The variable W is further defined as a variable W _important for calculating the average treatment effect.

In step S54, the average therapeutic effect of the target drug is calculated using the variables used to calculate the average therapeutic effect.

The framework proposed in this example can evaluate the effect of a target drug on the clinical outcome of efficacy/safety of multiple types of drugs. Define the average treatment effect (ATE) on the potential outcome Y of the drug of interest to be ATE=E(Y ₁ )−E(Y ₀ ), where E(Y _a ), ( a = 0, 1) indicates the probability that the outcome event would occur within the patient's complete follow-up period if all patients were assigned to treatment a in the simulated clinical trial. The potential outcomes are called counterfactuals because, for any given individual, only one of them is observed. In randomized controlled trials, the occurrence of outcome events between randomly assigned user cues and non-user cues is measured, and E(Y ₁ ) is directly estimated as E(Y|A=1). and E(Y ₀ ) can be estimated as E(Y|A=0). However, in real-world observational data, treatment assignment is not always random, so this example uses a double robust method to eliminate the influence of confounding factors to the maximum extent possible. In this example, alternative drugs are successively randomly selected from drugs used by patients with non-user cues in a restorative manner, and the average treatment effect (Average Treatment Effect, ATE) is calculated, and the calculation formula for ATE is defined as follows.

In step S541, an outcome event prediction model is constructed using the treatment method and the variables used to calculate the average treatment effect as input variables, and an estimated value of the conditional expected value of the outcome event is obtained.

を入力変数とするアウトカムイベント予測モデル

を使用し、Ｑ^０（Ａ，Ｗ）の値を結果Ｙの条件付き期待値の推定値

とする。このモデルの構築は、ロジスティック回帰等の任意の機械学習分類方法を使用して、Ａ＝０である場合の結果Ｙの条件付き期待値の推定値Ｑ^０（０，Ｗ）及びＡ＝１である場合の結果Ｙの条件付き期待値の推定値Ｑ^０（１，Ｗ）をそれぞれ計算することができる。 For treatment scheme A, a fitting estimation is performed for the conditional expectation value E(Y|A,W) of outcome Y under the given conditions of the important variable W _important , i.e., the value screened in the previous step. covariable

Outcome event prediction model with input variables

and let the value of Q ⁰ (A, W) be the estimate of the conditional expectation of the outcome Y

shall be. The construction of this model uses any machine learning classification method, such as logistic regression, to estimate the conditional expectation of the outcome Y when A= ⁰ and An estimate Q ⁰ (1, W) of the conditional expected value of the outcome Y for a certain case can be calculated respectively.

In step S542, the probability of treatment assignment is evaluated to obtain an estimate of treatment assignment.

Estimate the probability of treatment assignment P(A|W) and denote it as g(A,W), and the model can be used to estimate the probability of treatment assignment after fitting the model using any instrumental learning classification method such as logistic regression. An estimate of the probability g(1,W) of A=1 and an estimate g(0,W) of the probability of treatment assignment A=0 can also be calculated.

In step S543, an instruction disturbance variable is constructed using the estimated value of the treatment assignment.

The treatment assignment estimate g(A,W) is used to construct the indicated disturbance variable H(A,W).

Here, I is an indicator function, and when A=1, I(A=1)=1.

In step S544, a regression model of the outcome event and the indicated disturbance variable is constructed using the estimated value of the conditional expected value of the outcome event calculated by inputting the variables of the treatment method and the average treatment effect as an intercept term, Obtain estimated disturbance parameters.

を構築し、モデルをフィッティングした後に、指示外乱変数Ｈ（Ａ，Ｗ）の係数は推定外乱パラメータεである。 At this time, in this example, there is a deviation between Q ⁰ (A, W) and E (Y|A, W), and therefore it is necessary to correct Q (A, W) with the estimated disturbance parameter ε. be. As for how to use it, the value of Q ⁰ (A, W) is used as the intercept term, and the outcome event and indicated disturbance variable regression model, i.e.

After constructing and fitting the model, the coefficient of the indicated disturbance variable H(A,W) is the estimated disturbance parameter ε.

In step S545, the estimated value of the conditional expected value of the outcome event is updated using the estimated disturbance parameter and the indicated disturbance variable, and the average treatment effect is calculated.

Complete the update of Q ⁰ (A, W).

ATE can be calculated using the formula below.

where n' is the number of patients in the user queue and the number of patients currently using the selected alternative drug in the non-user queue, and i is the i-th record of the data set.

In step S55, steps S52 to S54 are repeated until the maximum number of tests is reached, the average effect of all the average therapeutic effects is calculated, and the confidence interval of the average effect is calculated to determine the effectiveness of the target drug. Determine safety.

In step S551, steps S52 to S54 are repeated until the maximum number of tests is reached, and the average effect of all the average treatment effects is calculated.

The maximum number of tests r is determined based on the number n _d of types of drugs taken by the patient in the original non-user queue.

の計算は以下のとおりである。

Once r trials are repeated with random selection of alternative drugs in the non-user cue, the final average effect of the target drug on the outcome measure

The calculation is as follows.

Here, ATE _i is the average treatment effect of the target drug and the i-th randomly selected alternative drug.

In step S552, a confidence interval of the average effect is calculated, and a signal with a left confidence interval larger than zero or a signal with a right confidence interval smaller than zero is obtained as a final positive signal, and the effectiveness of the target drug is determined. Determine safety.

を計算する必要があり、ここでσ_ＡＴＥはサブランダムシミュレーションＡＴＥの標準偏差である。アウトカム指標が有効性指標である場合、

のシグナルを最終的な陽性シグナルとして選択し、すなわち該薬物は対応する肯定的なアウトカムが発生すると見なし、アウトカム指標が安全性指標である場合、

のシグナルを最終的な陽性シグナルとして選択し、すなわち該薬物は対応する安全性アウトカムが発生すると見なす。 An additional 95% confidence interval is used to complete the causal effect between the drug and the outcome measure finally identified.

needs to be calculated, where σ _ATE is the standard deviation of the subrandom simulation ATE. If the outcome measure is an effectiveness measure,

is selected as the final positive signal, i.e. the drug is considered to produce the corresponding positive outcome, and the outcome indicator is a safety indicator;

is selected as the final positive signal, ie the drug is considered to produce the corresponding safety outcome.

The above are only preferred embodiments of the present invention and are not intended to limit the present invention, and the present invention may have various modifications and variations to those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims (6)

A high-throughput real-world evaluation method for drug efficacy and safety, comprising steps S1 to S5,
In step S1, real world data is acquired, integrated data coding is performed on the real world data, and after the integrated data coding, the minimum necessary data of the real world data is extracted to complete the construction of the dataset. death,
In step S2, a triplet mode is constructed that includes a concept set, an action, and a standard, the concept set is composed of one or more data codes, the action is composed of a series of actions or states, and the standard is representing the meaning of the level of efficacy/safety index, constructing a drug efficacy/safety index phenotype library according to the triplet mode, drug efficacy/safety term set, and clinical index;
In step S3, a target drug is selected, an indicator is selected from the efficacy/safety indicator phenotype library of the drug, a target drug-efficacy/safety indicator pair is obtained, and a target event, outcome event, and target Define date, outcome date, extract patient ID, target date and outcome date, configure a dataset for high-throughput signal primary screening,
The step S3 includes sub-steps S31 to S33,
In the sub-step S31, the first use of the target drug is set as a target event, one or more indicators selected by the user from the drug efficacy/safety indicator phenotype library are configured as an outcome indicator, and each of the above-mentioned The first occurrence of an outcome indicator is an outcome event,
In the sub-step S32, the date of occurrence of the target event is set as a target date, the date of occurrence of the outcome event is set as an outcome date,
In the sub-step S33, the target event and the outcome event are selected, the patients in the dataset in which both the target event and the outcome event occur are screened, and the patient ID, target date, and outcome date are extracted; Construct a dataset for high-throughput signal primary screening,
In step S4, high-throughput signal screening of the target drug-efficacy/safety indicator pair is performed on the data set of the high-throughput signal primary screening using event sequence symmetry analysis, and positive signals of the primary screening are detected. Acquired,
The step S4 includes sub-steps S41 to S44,
In the sub-step S41, the ordinal ratio is calculated using the ratio of the number of people for whom the outcome event occurred after the target event occurred and the number of people for whom the outcome event occurred before the target event occurred. calculate,
In the sub-step S42, an ineffective ordinal ratio is calculated, and the ineffective ordinal ratio is used to adjust the ordinal ratio in the sub-step S41 to obtain a modified ordinal ratio;
In the sub-step S43, a confidence interval of the modified ordinal ratio is calculated by calculating a confidence interval of the probability that the outcome event occurs after the target event occurs;
In the sub-step S44, a signal in which the lower limit of the confidence interval of the modified ordinal ratio is greater than 1 is determined to be a primary screening positive signal;
In step S5, a causal evaluation is performed on the positive signal of the primary screening to determine the effectiveness and safety of the target drug,
The step S5 includes sub-steps S51 to S55,
In the sub-step S51, patients in the data set in which the target event has occurred are screened for the target drug-efficacy/safety indicator pair corresponding to the primary screening positive signal, and the patients are screened in the user queue and Assign to the initial non-user queue,
In the sub-step S52, a type of drug is randomly selected as an alternative drug from among the drugs used by the patient in the initial non-user cue, and a non-user cue is constructed;
In the sub-step S53, the first use time of the target drug or the alternative drug is the registration time, the use of the target drug or the alternative drug is the treatment method, and whether or not the outcome event occurs is determined as the result; extracting baseline data before the patient enrolls, configuring the demographic information of the dataset and the baseline data as a covariate set, extracting features, screening and calculating an average treatment effect; Get the variables used in
The sub-step S53 specifically includes sub-steps S531 to S538,
In the sub-step S531, one covariate is selected from the covariable set, the covariate is designated as an importance covariate, the remaining variables in the covariable set are designated as remaining variables, and specifically, In the data set, the data structure is O=(W, A, Y), where W=W ₁ , W ₂ ,..., W _k is a set of covariates, including demographic information and Consisting of extracted baseline data, A is the treatment method, i.e. whether to use the target drug or an alternative drug, and Y is the outcome, i.e. whether the outcome event occurs or not. can be,

is the importance covariate whose importance is to be rated, and

are the remaining variables,
In the sub-step S532, an outcome event prediction model using the overall covariate as an input variable is constructed for the conditional expected value of the result under conditions given the importance covariate and the remaining variables. and obtain an estimate of the conditional expected value of the outcome, specifically, the importance covariate

, remaining variables

The conditional expectation value of the outcome Y under the given conditions

In other words, the outcome event prediction model uses the overall covariate W as an input variable.

and evaluate the importance of each covariate.

For,

The value of is the conditional expected value of the outcome Y

is an estimate of , using any machine learning classification method such as logistic regression,

Estimate of the conditional expectation of outcome Y if

as well as

Estimate of the conditional expectation of outcome Y if

Calculate each of
In the sub-step S533, the probability that the value of the importance covariate is 1 is estimated under the condition that the remaining variables are given, and the estimated value is obtained.

The importance covariate under the assumption that

The probability that is 1

Estimate

and after fitting the model using any instrumental learning classification method such as logistic regression, the remaining variables

The importance covariate under the assumption that

estimate of the probability that is 1

and the remaining variables

The importance covariate under the assumption that

estimate of the probability that is 0

Calculate,
In the sub-step S534, a variable importance indicating disturbance variable is constructed using the estimated value of the probability that the importance covariate is 1, and specifically,

Construct the variable importance indicating disturbance variable H _j by

Here, I is an indicator function,

If it is,

and
In the sub-step S535, the estimated value of the conditional expected value of the result is used as an intercept term, the variable importance indicating disturbance variable is used as an input variable, an outcome event regression model is constructed, and a variable importance disturbance parameter is obtained; in particular,

and the conditional expected value of the outcome Y

If there is a deviation between the variable importance disturbance parameter ε _j

How to fix and fix

A regression model of the outcome event and the disturbance variable indicating variable importance, with the value of as the intercept term, i.e.

After building and fitting the model, the variable importance indicates the disturbance variable

The coefficient of is the variable importance disturbance parameter ε _j ,
In the sub-step S536, the estimated value of the conditional expected value of the result is updated using the variable importance disturbance parameter and the variable importance indicating disturbance variable, and the variable importance is calculated. Specifically, In the following equation 8

Complete the update of

The variable importance is calculated by Equation 9 below,

where n is the number of patients in the dataset, i is the i-th record in the dataset,
In the sub-step S537, the standard deviation and confidence interval of the variable importance are calculated, specifically, in order to complete the variable screening, it is necessary to calculate the standard deviation of the variable importance, and the calculation formula is As shown in equation 10 below,

here,

, Var indicates the sample variance, and the 95% confidence interval of VI _j is

and
In the sub-step S538, the sub-steps S531 to S537 are repeated until each covariate of the covariate set is traversed, and the variables whose left and right confidence intervals of the standard deviation of variable importance do not include zero are Variables used to calculate the average treatment effect, specifically the left and right confidence intervals

Further, the variables are used as variables for calculating the average treatment effect,
In the sub-step S54, the average therapeutic effect of the target drug is calculated using the variables used to calculate the average therapeutic effect,
In the sub-step S55, the sub-step S52 to the sub-step S54 are repeated until the maximum number of tests is reached, the average effect of all the average treatment effects is calculated, and the confidence interval of the average effect is calculated. A high-throughput, real-world evaluation method for drug efficacy and safety, characterized by determining the efficacy and safety of a target drug. The sub-step S42 specifically includes sub-steps S421 to S423,
In the sub-step S421, the overall average probability that the outcome event occurs within a preset observation window after the target event occurs is calculated for the data set that performs high-throughput signal primary screening. ,
In the sub-step S422, a no-effect order ratio is calculated using the overall average probability,
The high-throughput real-time processing according to claim 1, characterized in that, in the sub-step S423, a ratio between the ordinal ratio in the sub-step S41 and the ineffective ordinal ratio is calculated to obtain a modified ordinal ratio. Methods for evaluating drug efficacy and safety in the world. The sub- step S51 specifically includes sub-steps S511 to S512,
In the sub-step S511, the patient in which the target event occurred in the data set is screened for the target drug-efficacy/safety index pair corresponding to the primary screening positive signal, and obtain a diagnosis of the patient, perform statistics on the patient based on the ICD-10 code, and set the main diagnosis with the largest number of patients as the basic diagnosis;
The method of claim 1, wherein the sub-step S512 assigns the patient to a user queue and an initial non-user queue. The sub- step S54 specifically includes sub-steps S541 to S545,
In the sub-step S541, an outcome event prediction model is constructed using the treatment method and the variables used to calculate the average treatment effect as input variables, and an estimated value of the conditional expected value of the outcome event is obtained;
The sub-step S542 evaluates the probability of treatment assignment to obtain an estimate of treatment assignment;
In the sub-step S543, an instruction disturbance variable is constructed using the estimated value of the treatment assignment;
In the sub-step S544, a regression model of the outcome event and the indicated disturbance variable is constructed using the estimated value of the conditional expected value of the outcome event calculated by inputting the variables of the treatment method and the average treatment effect as an intercept term. and obtain the estimated disturbance parameters,
According to claim 1, in the sub-step S545, the estimated value of the conditional expected value of the outcome event is updated using the estimated disturbance parameter and the indicated disturbance variable, and an average treatment effect is calculated. A high-throughput, real-world method for evaluating drug efficacy and safety. The sub- step S55 specifically includes sub-steps S551 to S552,
In the sub-step S551, repeat the sub-steps S52 to S54 until the maximum number of tests is reached, and calculate the average effect of all the average treatment effects,
In the sub-step S552, a confidence interval of the average effect is calculated, and a signal with a left confidence interval larger than zero or a signal with a right confidence interval smaller than zero is obtained as a final positive signal, and the effectiveness of the target drug is determined. 2. The high-throughput, real-world evaluation method for drug efficacy and safety according to claim 1, wherein the method comprises determining the efficacy and safety of a drug. A system based on the high-throughput real-world drug efficacy and safety evaluation method according to any one of claims 1 to 5 , comprising a data collection module, a high-throughput signal screening module, and a causal evaluation module. and a result display module,
the data collection module is used to obtain data from existing data and clean the data to complete the construction of the dataset;
The high-throughput signal screening module is used to convert data in the dataset into clinical events, perform high-throughput signal screening with event sequence symmetry analysis, and obtain a primary screening positive signal;
The causal evaluation module performs causal evaluation on the positive signal of the primary screening, and is used to determine the effectiveness and safety of the target drug,
A high-throughput real-world drug efficacy and safety evaluation system, wherein the result display module is used to display efficacy and safety results of the target drug.

Images