RUCAM scale automatic scoring and DILI analysis method based on interpretable artificial intelligence

Abstract

The invention provides a RUCAM scale automatic scoring and DILI analysis method based on interpretable artificial intelligence, and the method is characterized in that based on a liver index abnormal case, a time sequence method is adopted to analyze the sequence between the doctor's advice and the LIS index of a patient, information retrieval and slot grammar are adopted to carry out the semantic analysis of the disease course record of the patient, and the semantic analysis of the disease course record of the patient is carried out. Therefore, automatic scoring and DILI analysis of the RUCAM scale are realized, and an interpretable scoring basis is provided. According to the method, an interpretable artificial intelligence technology is adopted, automatic scoring is carried out on an RUCAM scale, a scoring basis is given, and causal relationship evaluation is carried out on DILI, so that the burden of manually reporting ADR is remarkably relieved.

Description

Technical Field

[0001] The present invention relates to the medical field, and in particular to an automatic scoring and DILI analysis method of the RUCAM scale based on explainable artificial intelligence. Background Art

[0002] In the process of drug use, liver damage caused by the drug itself and / or its metabolites or due to the hypersensitivity or reduced tolerance of special physical constitution to the drug is called drug-induced liver injury (DILI), also known as drug-induced liver disease. Clinically, it can manifest as various acute and chronic liver diseases. Mild cases can recover on their own after stopping the drug, while severe cases may be life-threatening and require active treatment and rescue. DILI can occur in healthy people with no history of liver disease or in patients with serious diseases; it can occur when the drug is overdosed or in normal dosage. There are more than 30,000 kinds of drugs and health products that are exposed in daily life, and more than 1,000 drugs that can cause DILI. Therefore, DILI has become a serious public health problem that cannot be ignored.

[0003] Causality assessment is required in the DILI diagnostic process, but the reliability of current causality assessment methods is not satisfactory or lacks external verification. The diagnosis of DILI is not only a scientific inference based on logic, but also an art that requires a comprehensive and detailed medical history and clinical and laboratory examination information to make a judgment. This highlights the importance of collecting detailed medical history of suspected drugs / concomitant medications, the occurrence / evolution of suspected DILI events, previous history of liver damage / liver disease, and concomitant diseases in the process of DILI diagnosis and differential diagnosis.

[0004] Causality assessment is an important and necessary step in the diagnostic process of DILI. However, the reality is that there is currently a lack of causality assessment methods with sufficient sensitivity, specificity, and accuracy.

[0005] In 1985, the pharmacovigilance department of Roussel Uclaf (RU) Pharmaceutical Company in Paris, France, launched a qualitative CAM that was not organ-specific to determine the causal relationship between adverse reactions and drug use. In 1988, RU Pharmaceutical Company organized a consensus meeting and improved and launched a CAM for acute cytolytic hepatitis based on the early CAM. In 1989, at the request and advocacy of the Council for International Organizations of Medical Sciences (CIOMS), an international consensus meeting was organized to discuss how to diagnose and treat acute DILI. Based on the above qualitative CAM, a liver-specific qualitative CAM was initially formed by adapting part of the qualitative assessment into assessment items specifically for DIL1, called the CIOMS scale. After that, based on the CIOMS scale, it took several years to weight each element and related details and assign appropriate scores. In 1993, the first liver-specific and quantitative DIL1 causal relationship assessment scale in the world was officially published, officially named RUCAM. Although the RUCAM scale itself has its flaws, it is still the causal relationship assessment method recommended by all current international guidelines because it can provide framework guidance.

[0006] However, in current practice, although the RUCAM evaluation method is used for evaluation, each drug used by the patient needs to be scored and evaluated, and each drug evaluation requires many parameters. If manual calculation is used, the workload is huge, time-consuming and labor-intensive. On the other hand, there will be problems of missing information or misjudgment based on experience in the manual calculation process. Summary of the invention

[0007] The present invention aims to overcome the above-mentioned defects by using explainable artificial intelligence technology to automatically score the RUCAM scale, provide a basis for scoring, and conduct a causal relationship assessment of DILI, so as to significantly reduce the burden of manual reporting of ADRs.

[0008] The present invention provides an automatic scoring and DILI analysis method for the RUCAM scale based on explainable artificial intelligence, which is characterized by: using a time series method to analyze the sequence between patient medical orders and LIS indicators, using information retrieval and slot grammar to perform semantic parsing on patient medical records, thereby realizing automatic scoring and DILI analysis of the RUCAM scale, and providing an explainable scoring basis.

[0009] The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence provided by the present invention is also characterized in that it comprises the following steps:

[0010] S1. Retrieve the patient's medical information and determine whether the patient's medical information contains conditions that trigger the automatic scoring of the RUCAM scale.

[0011] When "included", proceed to S2;

[0012] S2. Scoring is performed based on the RUCAM scale items, and the DILI analysis results and basis are given;

[0013] Among them, the disease course time series needs to be constructed in the process of determining whether the patient's diagnosis and treatment information contains the conditions that trigger the automatic scoring of the RUCAM scale;

[0014] The construction of the disease course time series is a disease course time series analysis for automatic scoring of the RUCAM scale, which requires obtaining LIS indicators and medical advice along the timeline, and then analyzing the time sequence, calculating the length of the time interval and information related to the timeline, and constructing the disease course time series based on this.

[0015] The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence provided by the present invention is also characterized in that the conditions triggering the automatic scoring of the RUCAM scale include: abnormal liver index conditions and suspected drug conditions.

[0016] The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence provided by the present invention is also characterized in that the method for constructing the disease course time series is as follows:

[0017] SA1. Extract LIS indicators;

[0018] SA2. Sort by time;

[0019] SA3. Determine whether there is an abnormality in the LIS item, and mark it if there is an abnormality;

[0020] SA4. Determine whether there is an abnormality in the liver index LIS group, and mark it if there is an abnormality;

[0021] SA5. Extract the doctor's orders and determine whether there are any suspected drugs. If there are any suspected drugs, mark them;

[0022] SA6. Combine LIS indicators and physician orders;

[0023] SA7. Sort LIS indicators and physician orders by time.

[0024] The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence provided by the present invention is also characterized in that the method for determining whether the patient's diagnosis and treatment information contains conditions that trigger the RUCAM scale automatic scoring is as follows:

[0025] SC1. Based on the time series of the course of disease, determine whether there are abnormal markers of the liver index LIS group;

[0026] When present, SC2 was performed;

[0027] SC2. Determine whether there are suspected drug markers from the date of admission to the date of abnormal liver indicators;

[0028] When present, it triggers the condition for automatic scoring of the RUCAM scale.

[0029] The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence provided by the present invention is also characterized in that the scoring based on the RUCAM scale items includes scoring item 1 based on time series analysis from medication to onset, scoring item 2 based on analysis of the course of the disease, scoring item 3 based on risk factors, scoring item 4 based on evaluation of concurrently used drugs, scoring item 5 based on exclusion of other reasons, scoring item 6 based on previous reports of liver damage caused by drugs, and scoring item 7 based on response to drug re-stimulation.

[0030] The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence provided by the present invention is also characterized in that before scoring each scoring item, DILI typing is required;

[0031] The method of DILI typing is as follows:

[0032] SD1. Retrieve ALT and ALP data;

[0033] SD2. Get the R value based on R = [ALT / ALT upper limit of normal range] / [ALP / ALP upper limit of normal range];

[0034] SD3. When R ≥ 5.0, it is marked as hepatocellular type;

[0035] When R≤2.0, it is marked as cholestatic type;

[0036] Others, labeled as mixed.

[0037] The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence provided by the present invention is also characterized in that the specific method for scoring based on the RUCAM scale items is as follows:

[0038] SE1. Get the score of scoring item 1;

[0039] SE1-1. Retrieve medication information and indicator abnormality information, and determine whether the indicator abnormality node is before "starting medication" or after "starting medication";

[0040] When the node is before “start taking medication”, it is judged as irrelevant;

[0041] When the node is "start taking medication", perform SE1-2;

[0042] SE1-2. Get the time of first taking the medicine and the time when the abnormality of the index began.

[0043] When the node is such as the time from medication to onset cannot be obtained, it is considered as unavailable for evaluation;

[0044] When the node is the time from medication availability to onset, SE1-2-1;

[0045] SE1-2-1. Calculate R value

[0046] When the node is R ≥ 5.0, perform SE1-2-1-1;

[0047] SE1-2-1-1. Determine whether the node with abnormal indicators is "initial treatment" or "subsequent treatment";

[0048] When the node is “initial treatment,” SE1-2-1-1-1 is performed;

[0049] SE1-2-1-1-1. Determine whether to stop taking the medicine when the indicators begin to become abnormal;

[0050] When the node is not stopped, perform SE1-2-1-1-1-1

[0051] SE1-2-1-1-1-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal;

[0052] When the node is 5-90 days, score item 1 will be increased by 2 points;

[0053] When the node is <5 days or >90 days, score item 1 is increased by 1 point;

[0054] When the node is in drug withdrawal, perform SE1-2-1-1-1-2;

[0055] SE1-2-1-1-1-2. Calculate the difference between the time of discontinuation of medication and the time when the indicator starts to become abnormal;

[0056] When the node is ≤15 days, score item 1 is increased by 1 point;

[0057] When the node is >15 days, perform SE1-2-1-1-1-2-1;

[0058] SE1-2-1-1-1-2-1. Determine whether the drug being used is a slowly metabolized chemical drug (amiodarone, leflunomide, clavulanate, etc.);

[0059] When the node is in use, score item 1 is increased by 1 point;

[0060] When a node is unused, it is judged as irrelevant;

[0061] When the node is “Subsequent treatment,” perform SE1-2-1-1-2;

[0062] SE1-2-1-1-2. Determine whether to stop taking the medicine when the indicators start to become abnormal;

[0063] When the node is not discontinued, perform SE1-2-1-1-2-1;

[0064] SE1-2-1-1-2-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal;

[0065] When the node is 1-15 days, score item 1 will be increased by 2 points;

[0066] When the node is >15 days, score item 1 is increased by 1 point;

[0067] When the node is in drug withdrawal, perform SE1-2-1-1-2-2;

[0068] SE1-2-1-1-2-2. Calculate the difference between the time of discontinuation of medication and the time when the indicator starts to become abnormal;

[0069] When the node is ≤15 days, score item 1 is increased by 1 point;

[0070] When the node is >15 days, perform SE1-2-1-1-2-2-1;

[0071] SE1-2-1-1-2-2-1. Determine whether the drug being used is a slowly metabolized chemical drug;

[0072] When the node is in use, score item 1 is increased by 1 point;

[0073] When a node is unused, it is judged as irrelevant;

[0074] When the node is R<5.0, perform SE1-2-1-2;

[0075] SE1-2-1-2. Determine whether the node with abnormal indicators is "initial treatment" or "subsequent treatment";

[0076] When the node is “initial treatment,” perform SE1-2-1-2-1;

[0077] SE1-2-1-2-1. Determine whether to stop taking the medicine when the indicators begin to become abnormal;

[0078] When the node is not stopped, perform SE1-2-1-2-1-1

[0079] SE1-2-1-2-1-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal;

[0080] When the node is 5-90 days, score item 1 will be increased by 2 points;

[0081] When the node is <5 days or >90 days, score item 1 is increased by 1 point;

[0082] When the node is in drug withdrawal, perform SE1-2-1-2-1-2;

[0083] SE1-2-1-2-1-2. Calculate the difference between the time of discontinuation of medication and the time when the index begins to become abnormal;

[0084] When the node is ≤30 days, score item 1 is increased by 1 point;

[0085] When the node is >30 days, perform SE1-2-1-2-1-2-1;

[0086] SE1-2-1-2-1-2-1. Determine whether the drug being used is a slowly metabolized chemical drug;

[0087] When the node is in use, score item 1 is increased by 1 point;

[0088] When a node is unused, it is judged as irrelevant;

[0089] When the node is “Subsequent treatment”, perform SE1-2-1-2-2;

[0090] SE1-2-1-2-2. Determine whether to stop taking the medicine when the indicators begin to become abnormal;

[0091] When the node is not discontinued, perform SE1-2-1-2-2-1;

[0092] SE1-2-1-2-2-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal;

[0093] When the node is 1-90 days, score item 1 plus 2 points;

[0094] When the node is >90 days, score item 1 is increased by 1 point;

[0095] When the node is in drug withdrawal, perform SE1-2-1-2-2-2;

[0096] SE1-2-1-2-2-2. Calculate the difference between the time of discontinuation of medication and the time when the indicator starts to become abnormal;

[0097] When the node is ≤30 days, score item 1 is increased by 1 point;

[0098] When the node is >30 days, perform SE1-2-1-2-2-2-1;

[0099] SE1-2-1-2-2-2-1. Determine whether the drug being used is a slowly metabolized chemical drug;

[0100] When the node is in use, score item 1 is increased by 1 point;

[0101] When a node is unused, it is considered irrelevant.

[0102] SE2. Get the score of scoring item 2;

[0103] SE2-1. Retrieve medication information and abnormal indicator information to determine whether to stop taking the medication;

[0104] When the node is still in use and there is no conclusion, score item 2 is increased by 0 points;

[0105] When the node is drug withdrawal, perform SE2-1-1;

[0106] SE2-1-1. Calculate the R value;

[0107] When the node is R ≥ 5.0, perform SE2-1-1-1;

[0108] SE2-1-1-1. Difference between ALT peak and upper limit of normal;

[0109] If the node drops by ≥50% within 8 days, 3 points will be added to scoring item 2;

[0110] If the node is reduced by ≥50% within 30 days, 2 points will be added to scoring item 2;

[0111] If there is no relevant data for a node or it decreases by ≥50% after 30 days, score item 2 will be increased by 0 points;

[0112] If the node decreases by <50% or increases again after 30 days, score item 2 will be increased by -2 points;

[0113] When the node is R<5.0, perform SE2-1-1-2;

[0114] SE2-1-1-2. Difference between peak ALP or TBil value and upper limit of normal;

[0115] If the node drops by ≥50% within 180 days, 2 points will be added to scoring item 2;

[0116] If the node drops by less than 50% within 180 days, score item 2 will be increased by 1 point;

[0117] When the node is unchanged, rising, or has no data, score item 2 is increased by 0 points.

[0118] SE3. Get the score of scoring item 3;

[0119] SE3-1. Retrieve the patient's medical history information to determine whether the patient has drinking behavior (current drinking volume: female>2 times / day, male>3 times / day, about 10g ethanol / time);

[0120] When the node is "yes", score item 3 is increased by 1 point;

[0121] When the node is "No", proceed to SE3-2;

[0122] SE3-2. Retrieve patient information and determine whether the patient is pregnant;

[0123] When “No”, score item 3 is added with 0 points;

[0124] If “yes”, proceed to SE3-2-1;

[0125] Perform SE3-2-1. Calculate the R value;

[0126] When the node is R<5.0, score item 3 is increased by 1 point;

[0127] When the node is R≥5.0, score item 3 is increased by 0 points;

[0128] SE3-3. Retrieve the patient's age information and determine whether the patient is over 55 years old;

[0129] When the node is age ≥ 55, score item 3 is increased by 1 point;

[0130] When the node is age < 55, score item 3 is added with 0 points;

[0131] SE4. Get the score of scoring item 4;

[0132] SE4-1. Retrieve the patient's medication information and determine:

[0133] If there is no or the time of concomitant medication use does not match the time of onset, 0 points will be added to scoring item 4;

[0134] If the duration of concomitant medication use is consistent with the onset of illness, -1 point will be added to item 4;

[0135] If the concomitant medication is known to be hepatotoxic and the duration of use is consistent with the onset of the disease, -2 points will be added to score item 4;

[0136] If there is evidence of liver damage caused by concomitant medication (re-medication reaction or valuable test), score item 4 plus -3 points;

[0137] SE5. Get the score of scoring item 5;

[0138] Other possible causes of liver disease are divided into 2 groups:

[0139] Group I (7 types of causes):

[0140] HAV infection: Anti-HAV-IgM

[0141] HBV infection: HBsAg, anti-HBc-IgM, HBV DNA

[0142] HCV infection: Anti-HCV, HCV RNA

[0143] HEV infection: anti-HEV-IgM, anti-HEV-IgG, HEV RNA

[0144] Hepatobiliary ultrasound imaging / liver vascular color Doppler imaging / endovascular ultrasound examination / CT / MRC

[0145] Alcohol intoxication (AST / ALT ≥ 2)

[0146] Recent history of acute hypotension (especially in the setting of underlying heart disease)

[0147] Group II (5 types of causes):

[0148] Combined with sepsis, metastatic malignant tumors, autoimmune hepatitis, chronic hepatitis B or C, primary biliary cholangitis or primary sclerosing cholangitis, hereditary liver disease, etc.

[0149] CMV infection: anti-CMV-IgM, anti-CMV-IgG, CMV-PCR

[0150] EBV infection: anti-EBV-IgM, anti-EBV-IgG, EBV-PCR

[0151] HSV infection: anti-HSV-IgM, anti-HSV-IgG, HSV-PCR

[0152] VZV infection: anti-VZV-IgM, anti-VZV-IgG, VZV-PCR

[0153] SE5-1. Retrieve the patient's medical history information and determine:

[0154] If all causes in groups I and II can be reasonably excluded, 2 points are added to item 5;

[0155] If all seven causes in group I can be ruled out, but the conditions in group II cannot be ruled out, 1 point is added to item 5;

[0156] If only 5 to 6 causes in group I can be excluded, score item 5 plus 0 points;

[0157] If there are less than 5 causes that can be excluded in group I, 2 points will be subtracted from item 5;

[0158] If the presence of another liver disease is "highly likely", subtract 3 points from item 5;

[0159] SE6. Get the score of scoring item 6;

[0160] SE5-1. Collect reports on previous liver damage caused by drugs and determine:

[0161] If there is a report of hepatotoxicity in the product description, 2 points will be added to the score item 5;

[0162] If there is literature report but no relevant information in the product description, 1 point will be added to the score item 5;

[0163] There were no reports of hepatotoxicity, and the score item 5 plus 0 points;

[0164] SE7. Get the score of scoring item 7;

[0165] SE7-1. Retrieve the patient's medical history information to determine the reaction to medication:

[0166] When the node is R ≥ 5.0,

[0167] If ALT is below 5ULN before medication, and ALT doubles after medication / herbal medicine is used again, 3 points will be added to score item 7;

[0168] If the drug / herbal medicine used in the first reaction is given again and ALT doubles, 1 point is added to score item 7;

[0169] Under the same conditions as the first use of the drug, if ALT is elevated but below ULN, score item 7 is reduced by 2 points;

[0170] In other cases, 0 points will be added to scoring item 7;

[0171] When the node is R<5.0,

[0172] If the ALP level is below 2ULN before medication, and doubles after medication / herbal medicine is used again, add 3 points to item 7;

[0173] If the drug / herbal medicine used in the first reaction is given again and the ALP level doubles, 1 point will be added to item 7;

[0174] Under the same conditions as the first medication, if ALT is elevated but below ULN, score item 7 is subtracted by 2 points;

[0175] In other cases, add 0 points to scoring item 7.

[0176] The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence provided by the present invention is also characterized in that the analysis method of DILI analysis:

[0177] When the RUCAM scale is scored, 0 or less means that the drug is “excluded” as the cause of liver injury;

[0178] When the RUCAM scale score is 1 to 2 points, it means “impossible”;

[0179] When the RUCAM scale score is 3 to 5 points, it means “possible”;

[0180] When the RUCAM scale score is 6 to 8 points, it means “very likely”;

[0181] A RUCAM score greater than 8 indicates “high probability”.

[0182] The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence provided by the present invention is also characterized in that the method of semantic analysis of patient medical records using information retrieval and slot grammar is as follows:

[0183] SE1. To meet the needs of RUCAM scale scoring, information retrieval was performed in the content field based on the search terms;

[0184] If "yes", proceed to SE2;

[0185] If "not present", repeat SE1;

[0186] SE2. Determine whether the search term is in the title;

[0187] When "yes", filter and repeat SE1;

[0188] When “No”, proceed to SE3;

[0189] SE3. Perform sentence division and slot grammar parsing. DETAILED DESCRIPTION

[0190] The present invention can be implemented in many ways and can have various embodiments, so each specific embodiment will be illustrated and described. However, this is not intended to limit the present invention to a specific implementation, but should be understood to include all changes, equivalents and even substitutes that fall within the concept and technical scope of the present invention.

[0191] Example 1 Data and Methods

[0192] 1.1 Data

[0193] The patient cases come from a tertiary comprehensive hospital and are divided into two parts.

[0194] Among them: 1) Cases of patients with abnormal liver indicators collected from December 30, 2018 to September 10, 2023, including LIS, medical advice, medical records and other information, totaling 3,896 cases (3,375 of which triggered automatic scoring). On the one hand, 3 typical cases were selected to test the automatic scoring of the RUCAM scale and DILI analysis based on the explainable artificial intelligence method, and the feasibility and credibility of the method were evaluated. On the other hand, the RUCAM scale was automatically scored and DILI analyzed for all cases, and the analysis results based on big data were given. The usability of the method was evaluated on big data, and typical cases were analyzed.

[0195] 2) From June 2023 to May 2024, 9 DILI cases successfully reported by a tertiary comprehensive hospital were automatically scored and analyzed using the RUCAM scale based on an explainable artificial intelligence method. The results were compared with the manually reported results, and the practicality of the method was evaluated.

[0196] 1.2 Methods

[0197] Including trigger conditions for automatic scoring of RUCAM scale, time series analysis method for automatic scoring of RUCAM scale, semantic analysis method for medical records for automatic scoring of RUCAM scale, scoring details of RUCAM scale, automatic scoring of RUCAM scale and DILI analysis algorithm based on explainable artificial intelligence, as follows:

[0198] 1.2.1 Trigger conditions for automatic scoring of the RUCAM scale

[0199] The conditions that trigger the automatic scoring of the RUCAM scale include abnormal liver index conditions and suspected drug conditions. The following abnormal liver index conditions must be met, and more than one suspected drug must have been taken before the abnormal liver index.

[0200] 1.2.1.1 Abnormal liver index conditions: Abnormal liver index conditions are alanine aminotransferase (ALT)>120umol / L.

[0201] 1.2.1.2 Suspected drug conditions: A tertiary general hospital manually compiled a list of suspected liver injury drugs based on drug instructions, literature reports, and whether the hospital is using them, totaling 774 drugs. Among them, a small number of drugs have the same name, and after removing the drugs with the same name, there are 739 drugs. For example: gemcitabine hydrochloride for injection, floxuridine for injection, serotonin capsules, actinomycin for injection, teniposide injection, etc.

[0202] 1.2.2 Time series analysis method for disease course automatic scoring of RUCAM scale

[0203] It includes the content and purpose of disease course time series analysis, disease course time series construction algorithm, and disease course time series query algorithm.

[0204] 1.2.2.1 Content and purpose of disease course time series analysis:

[0205] The contents of the time series analysis of the course of disease include: LIS indicators and doctor's orders. LIS indicators include AST, ALT, ALP, etc. Doctor's orders include the use of the above 739 suspected liver injury drugs. LIS indicators use the date and time of recording. Doctor's orders use the time when the doctor's orders are issued, and long-term doctor's orders are issued according to the number of days of medication.

[0206] The purpose of disease course time series analysis is to expand LIS indicators and medical advice along the timeline in order to analyze the temporal sequence and calculate the length of time intervals. For example, when a patient's liver indicators are abnormal, it is necessary to check whether the patient has taken suspected drugs from the beginning of the current hospitalization to the time point when the liver indicators are abnormal, whether the suspected drugs have been discontinued, and whether the liver indicators have returned to normal after the suspected drugs have been discontinued.

[0207] 1.2.2.2 Algorithm for constructing disease course time series

[0208]

[0209]

[0210] The algorithm for constructing the disease course time series is shown in Algorithm 1.1.

[0211] Among them: Input: liver index LIS results, including: ALT, TBil, ALP, etc. Medical orders include the above 739 suspected liver damage drugs. Output: LIS and medical orders arranged from front to back according to the timeline.

[0212] Row 1: Read in the LIS results and arrange them in timeline. According to the system configuration, mark whether the LIS items are abnormal. For example, whether ALT is greater than 120umol / L. Row 2: According to the abnormal liver index trigger conditions, if ALT, ALP, and TBIl are abnormal, mark whether the LIS group is abnormal. If abnormal, it meets the abnormal liver index trigger conditions of the RUCAM scale automatic scoring. Row 3: Read in the doctor's order. Row 4: Combine the LIS results and the doctor's order. Row 5: Arrange the LIS results and the doctor's order from front to back according to the timeline. See the time series of the course of three hospitalized patients given in Table 1 below.

[0213] Table 1. Time series of the course of illness of three patients with hospitalization IDs J83295, H61451, and K35375

[0214] Hospitalization ID date Hospitalization ID date Hospitalization ID date J83295 2022-06-16 H61451 2020-07-21 K35375 2022-11-15 Suspected drugs Suspected drugs Normal indicators 2022-11-15 Normal indicators 2022-06-17 Abnormal indicators 2020-07-31 Suspected drugs Suspected drugs Suspected drugs Abnormal indicators 2022-11-16 Normal indicators 2022-06-26 Suspected drugs Suspected drugs Abnormal indicators 2022-11-19 Normal indicators 2022-07-06 none Suspected drugs Abnormal indicators 2022-11-25 Normal indicators 2022-07-07 none Suspected drugs Abnormal indicators 2022-11-28 Abnormal indicators 2022-07-14 Suspected drugs Normal indicators 2022-07-25 Suspected drugs Normal indicators 2022-08-02 Suspected drugs Normal indicators 2022-08-13 Suspected drugs

[0215] 1.2.2.3 Disease course time series query algorithm

[0216] It includes the RUCAM scale's medication-to-onset time query algorithm and the RUCAM scale's disease course query algorithm.

[0217] in:

[0218] RUCAM scale scoring item 1 - time from medication to onset. This scoring item is divided into two parts: from the start of medication. Algorithm 1.2 for starting medication. (See 1.2.4 for specific scoring methods)

[0219] Among them, input: completed LIS and doctor's order time series. Output: whether to trigger the automatic scoring of RUCAM scale, and the score of the scoring item after triggering.

[0220] Line 1: First check whether the LIS group indicators are abnormal? That is, whether the liver indicator abnormality condition is triggered? Line 2: If not, return false. Line 3: If satisfied. Then check whether suspected drugs have been taken during the period from the date of admission to the abnormal liver indicators. Line 4: If not, return false. Line 5: If suspected drugs have been taken, trigger the RUCAM scale automatic scoring. Line 6: Return the score and scoring basis of RUCAM scale scoring item 1-from taking medicine to onset time.

[0221]

[0222] RUCAM scale scoring item 2 - course of disease. Taking cholestatic or mixed type as an example, see algorithm 1.3. (See 1.2.4 for specific scoring methods)

[0223] Input: Completely constructed LIS and doctor's order time series Output: RUCAM scale scoring item 2 - score of disease course.

[0224] Line 1: Check if the ALP index is abnormal, that is, beyond ULN. Line 2: If not, return 0 points. Line 3: If yes, calculate the difference between the abnormal value and ULN, and find out if there is a normal point for the index? Line 5: If no, return 0 points. Line 6: If yes, calculate according to the scoring criteria, return the score of the course of disease and the basis for the scoring.

[0225]

[0226] 1.2.3 Semantic analysis method of medical records for automatic scoring of RUCAM scale

[0227] It includes information retrieval for medical records, slot grammar, and semantic analysis methods of medical records based on information retrieval and slot grammar.

[0228] 1.2.3.1 Information retrieval for medical records

[0229] Information retrieval is the main way for users to search and obtain information. It is a method and means of finding information. Information retrieval in a narrow sense refers only to information search. That is, the user uses a certain method and retrieval tools to find the required information from the information collection according to the needs. Information retrieval in a broad sense is the process of processing, sorting, organizing and storing information in a certain way, and then accurately finding relevant information according to the specific needs of the information user. It is also called information storage and retrieval. In general, information retrieval refers to information retrieval in a broad sense.

[0230] The medical record includes the following four fields: patient hospitalization ID, title, content, and record date and time.

[0231] The title fields include: admission diagnosis and signs, brief condition, attending schedule_current diagnosis, patient name, gender, age, admission diagnosis, discharge diagnosis, etc. The title field can be used to assist in retrieval. For example: to retrieve whether the patient is pregnant? The field title "pregnancy history" needs to be excluded because this field indicates that the patient has been pregnant, not the current state.

[0232] The content field is the main field for information retrieval. For the following admission diagnosis and discharge diagnosis content, taking the keyword "tumor" as an example, the following content can be searched. However, to understand whether the patient has a tumor, semantic analysis based on slot grammar is required.

[0233] Admission diagnosis: 1. Sigmoid colon tumor; 2. Ascending colon polyp

[0234] Discharge diagnosis: 1. Rectal malignancy (tubular adenoma at the junction of the rectosigmoid colon, local high-grade intraepithelial neoplasia); 2. Ascending colon polyp (tubular adenoma); 3. Right cerebral artery occlusion and cerebral infarction; 4. Vasospasm; 5. Atelectasis; 6. Thrombocytosis; 7. Duodenal diverticulum;

[0235] 1.2.3.2 Slot grammar for medical records

[0236] Slot Grammar is a grammar used to process information vocabulary, mainly used in fields such as natural language processing and machine translation. The concept of slot grammar can be traced back to early natural language processing research, especially in machine translation and natural language understanding. It is used to decompose complex natural language text into structured information that is easier to process, thereby improving processing efficiency and accuracy. The basic concept is to map words or phrases in a sentence to predefined slots, which are used to represent specific semantic information. Slot grammar simplifies the task of natural language processing by defining a series of slots to capture key information in the language.

[0237] The basic principle of slot grammar is to map words or phrases in a sentence to predefined slots. These slots represent different semantic units, such as time, place, person, etc. In this way, natural language text is decomposed into a series of slots, each of which contains specific information. Implementing slot grammar usually involves the following steps: 1) Define slots: Define a series of slots according to the language and tasks that need to be processed. 2) Map vocabulary: Map words or phrases in a sentence to the corresponding slots. 3) Structured information: Organize the mapped slots into structured information representations for further processing and analysis. In short, slot grammar is an effective natural language processing method that simplifies the task of natural language processing and improves processing efficiency and accuracy by decomposing natural language text into structured information units.

[0238] By observing a large number of medical records, the slots studied in this paper include: 1) Disease symptoms, such as tumors, edema, alcoholism, etc. 2) Body parts: liver, lower limbs, both lower limbs, left eye, right eye, etc. 3) Negative words: not seen, not reached, no, no, etc. 4) Modifiers: superficial, severe, etc.

[0239] The vocabulary mapping rules include: masculine: +, ↑, masculine. Negative: -, ↓, feminine.

[0240] Structured information includes natural language and LIS. 1) Natural language. For example, "not addicted to smoking and drinking" is structured as follows: disease symptoms: smoking and drinking; body parts: NULL; negation: yes; modifier: NULL. It can be judged that the patient does not have a long-term drinking habit. 2) LIS. For example: Anti-HAV-IgM (+). LIS item: Anti-HAV-IgM; LIS value: positive. It can be judged that the patient's Anti-HAV-IgM is positive.

[0241] 1.2.3.3 Semantic Analysis Algorithm of Medical Records Based on Information Retrieval and Slot Grammar

[0242] The medical record semantic analysis algorithm is shown in Algorithm 1.4.

[0243]

[0244] Input: patient's medical history, search term list, RUCAM scoring item (e.g. scoring item 5 - exclude other causes) Output: score of RUCAM scoring item.

[0245] Among them, line 1: In view of the need for RUCAM scale scoring, information retrieval is performed in the content field based on the search terms. Line 2: If it does not exist, continue. Line 3: If it exists, check whether filtering is required based on the title field. Line 4: If filtering is required, continue. Line 5: If filtering is not required, sentence segmentation is performed and slot grammar parsing is performed. Line 6: If the parsing result is negative, continue. Line 7. If the parsing result is positive, score is performed based on the RUCAM scale items and the score is returned.

[0246] 1.2.4. RUCAM scale scoring details

[0247] According to the requirements of the RUCAM scale, seven items need to be scored. Among them, the seventh item "drug re-use reaction" is difficult to conduct and is no longer considered in this study. Only the remaining six items are scored.

[0248] Before scoring, DILI typing is required, including hepatocellular, cholestatic, or mixed typing. The calculation method of R value is shown in formula (1). Among them, the ULN of alanine aminotransferase ALT is 120umol / L, and the ULN of alkaline phosphatase ALP is 125umol / L.

[0249] R = [ALT / ALT upper limit of normal range] / [ALP / ALP upper limit of normal range] (1)

[0250] When R≥5.0, it is hepatocellular type. When R≤2.0, it is cholestatic type. Otherwise, it is mixed type.

[0251] 1.2.4.1. Scoring Item 1: Time from medication to onset of illness

[0252] This scoring item involves onset and initial treatment, which are defined as follows: Onset: abnormal indicators. That is, AST aspartate aminotransferase in LIS>60U / L, and ALT alanine aminotransferase>120umol / L. Initial treatment: the first time the suspected drug was taken during this hospitalization. This scoring item includes "starting from taking the drug" and "starting from stopping the drug".

[0253] (1) From the beginning of medication. From the beginning of medication, there are two situations: prompt and suspicious. 1) Prompt. First treatment. Also written as first medication. When the index is abnormal (instead of onset), find the duration from the first use of the suspected drug during this hospitalization to the abnormal index as the first medication-onset time. Case 1: If the first medication-onset time is 5-90 days, add 2 points. Case 2: If the first medication-onset time is more than 90 days, check the subsequent medication-onset time. This time is the duration of subsequent medication from the abnormal index back. If the subsequent medication-onset time is 1-90 days (cholestatic type and mixed type) or 1-15 days (hepatocellular type), add 2 points. Case 3: If the first medication-onset time is less than 5 days, check whether there is subsequent medication. If so, add 2 points. Others: If none of the above three situations are met, perform the following analysis. 2) Suspicious. If the initial treatment time is greater than 90 days or less than 5 days, 1 point will be added.

[0254] (2) Starting from drug discontinuation, the score is 0.

[0255] 1.2.4.2. Scoring Item 2: Disease Course

[0256] Taking cholestatic or mixed type as an example: First, obtain the ALP value of the abnormal index and calculate the difference between it and the upper limit of normal ULN. Secondly, starting from the time point of abnormal index, check whether there is a time point when the index is normal. If it exists, trace back from the time point when the index is normal to check the time of drug withdrawal (the most recent time the drug was discontinued before this treatment). There are three situations: 1) Prompt. If the difference between the ALP value exceeding the normal index and the upper limit of normal decreases by more than 50% within 180 days, add 2 points. 2) Suspicious. If the difference between the ALP value exceeding the normal index and the upper limit of normal decreases by less than 50% within 180 days, add 1 point. If not applicable, subtract 1 point. 3) No conclusion. If the difference between the ALP value exceeding the normal index and the upper limit of normal remains unchanged, increases, or there is no data, add 0 points. It should be noted that since "highly suggestive" (+3) and "opposite to the drug effect" are not applicable, they are not considered in this embodiment.

[0257] Hepatocellular type: R ≥ 5.0;

[0258] 1.2.4.3. Scoring Item 3: Risk Factors

[0259] Risk factors include: alcohol, pregnancy, and age.

[0260] Among them: 1) Alcohol or pregnancy. For cholestatic or mixed , if there is excessive drinking or pregnancy, add 1 point. Hepatocellular type does not consider pregnancy, only drinking.

[0261] 2) Age. Age was taken from the patient's hospitalization record. If the age was ≥55, 1 point was added. Otherwise, 0 point was added.

[0262] 1.2.4.4. Scoring Item 4: Concomitant Medication

[0263] There are four situations:

[0264] (1) No concomitant medication or the time of concomitant medication use does not match the time of onset. Add 0 points.

[0265] (2) If the duration of concomitant medication use matches the onset of disease, but it is unknown whether the concomitant medication is hepatotoxic, deduct 1 point.

[0266] (3) If the concomitant medication is known to be hepatotoxic and the duration of use is consistent with the onset of disease, 2 points will be deducted.

[0267] (4) If further evidence such as restimulation test or some confirmatory biomarker test shows that the drug can cause liver damage, 3 points will be added.

[0268] 1.2.4.5. Scoring Item 5: Other Reasons for Exclusion

[0269] Divided into two groups:

[0270] (1) Group I. 1) HAV infection. Anti-HAV-IgM or anti-HAV-IgM, HAV-IgM. 2) HBV infection. HBsAg, anti-HBc-IgM or anti-HBc-IgM, HBV DNA. 3) HCV infection. Anti-HCV or anti-HCV, HCV, HCV RNA. 4) HEV infection. Anti-HEV-IgM or anti-HEV-IgM, anti-HEV-IgG or anti-HEV-IgG, HEV-IgM, HEV-IgG, HEV RNA. 5) Hepatobiliary ultrasound imaging, hepatic vascular color Doppler imaging, intracavitary ultrasound, CT, MRC. The analysis is limited to the fields containing "[Test results]" in the title of the medical record. 6) Alcoholism, and AST / ALT ≥ 2. First, calculate whether AST / ALT ≥ 2 based on the LIS index. If "no", it is not satisfied. If "yes", check whether the word "alcohol poisoning" appears in the medical history. If it does, it is satisfied; if it does not appear, it is not satisfied. 7) Acute hypotension (especially when there is underlying heart disease)

[0271] (2) Group II. 1) Patients with concurrent sepsis, metastatic malignant tumors, autoimmune hepatitis, chronic hepatitis B or C, primary biliary cholangitis or primary sclerosing cholangitis, hereditary liver disease, etc. 2) CMV infection.

[0272] Anti-CMV-IgM or anti-CMV-IgM, anti-CMV-IgG or anti-CMV-IgG, CMV-IgM, CMV-IgG, CMVPCR.

[0273] 3) EBV infection: Anti-EBV-IgM or anti-EBV-IgM, anti-EBV-IgG or anti-EBV-IgG, EBV-IgM, EBV-IgG, EBV PCR.

[0274] 4) HSV infection. Anti-HSV-IgM or anti-HSV-IgM, anti-HSV-IgG or anti-HSV-IgG, HSV-IgM, HSV-IgG, HSV PCR. 5) VZV infection. Anti-VZV-IgM or anti-VZV-IgM, anti-VZV-IgG or anti-VZV-IgG, VZV-IgM, VZV-IgG, VZV PCR.

[0275] If all conditions in Group I and Group II can be reasonably excluded, 2 points are added. If all 6 factors in Group I can be excluded, but conditions in Group II cannot be excluded, 1 point is added. If only 4 to 5 factors in Group I can be excluded, 0 points are added. If less than 4 factors in Group I can be excluded, 2 points are subtracted. If the presence of another liver disease is found to be "highly likely", 3 points are subtracted.

[0276] 1.2.4.6. Scoring Item 6: Reports of previous liver injury caused by medication

[0277] If the drug label states that the drug has the potential to cause DILI, 2 points are added. If the drug label does not state this, but there are published case reports of liver injury caused by this drug, 1 point is added. If the adverse liver effects of the drug are unknown, 0 points are added.

[0278] 1.2.4.7. Scoring Item 7: Response to drug restimulation

[0279] In the case of inadvertent exposure or deliberate rechallenge, if the drug alone causes a doubling of the ALT level (for cases of hepatocellular injury) or a doubling of the ALP or TBILI level (for cases of mixed or cholestatic liver injury), a score of 3 is assigned. If the drug is rechallenge during acute liver injury and a doubling of the ALT, ALP, or TBILI level occurs, a score of 1 is assigned. If rechallenge of a drug after recovery from the initial liver injury does not cause an increase in ALT, ALP, or TBILI above the ULN, a score of -2 is assigned. No score (0) is assigned if no drug rechallenge or reexposure occurs.

[0280] The specific algorithm is as follows:

[0281] SE1. Get the score of scoring item 1;

[0282] SE1-1. Retrieve medication information and indicator abnormality information, and determine whether the indicator abnormality node is before "starting medication" or after "starting medication";

[0283] When the node is before "start taking medication", it is judged as irrelevant; when the node is after "start taking medication", SE1-2 is performed;

[0284] SE1-2. Get the time of first taking the medicine and the time when the abnormality of the index began.

[0285] When the node is such that the time from taking medication to onset cannot be obtained, it is considered as unavailable for evaluation; when the node is such that the time from taking medication to onset is available, it is considered as SE1-2-1;

[0286] SE1-2-1. Calculate R value

[0287] When the node is R ≥ 5.0, perform SE1-2-1-1;

[0288] SE1-2-1-1. Determine whether the node with abnormal indicators is "initial treatment" or "subsequent treatment"; when the node is "initial treatment", perform SE1-2-1-1-1;

[0289] SE1-2-1-1-1. Determine whether to stop taking the medicine when the indicators begin to become abnormal;

[0290] When the node is not discontinued, perform SE1-2-1-1-1-1;

[0291] SE1-2-1-1-1-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal;

[0292] When the node is 5-90 days, add 2 points to scoring item 1; when the node is <5 days or >90 days, add 1 point to scoring item 1; when the node is drug withdrawal, perform SE1-2-1-1-1-2;

[0293] SE1-2-1-1-1-2. Calculate the difference between the time of discontinuation of medication and the time when the indicator starts to become abnormal;

[0294] When the node is ≤15 days, add 1 point to scoring item 1; when the node is >15 days, perform SE1-2-1-1-1-2-1;

[0295] SE1-2-1-1-1-2-1. Determine whether the drug being used is a slowly metabolized chemical drug (amiodarone, leflunomide, clavulanate, etc.);

[0296] When the node is used, score item 1 is increased by 1 point; when the node is not used, it is judged as irrelevant; when the node is "subsequent treatment", SE1-2-1-1-2 is performed;

[0297] SE1-2-1-1-2. Determine whether to stop taking the medicine when the indicators start to become abnormal;

[0298] When the node is not discontinued, perform SE1-2-1-1-2-1;

[0299] SE1-2-1-1-2-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal;

[0300] When the node is 1-15 days, add 2 points to scoring item 1; when the node is >15 days, add 1 point to scoring item 1; when the node is drug withdrawal, perform SE1-2-1-1-2-2;

[0301] SE1-2-1-1-2-2. Calculate the difference between the time of discontinuation of medication and the time when the indicator starts to become abnormal;

[0302] When the node is ≤15 days, add 1 point to scoring item 1; when the node is >15 days, perform SE1-2-1-1-2-2-1;

[0303] SE1-2-1-1-2-2-1. Determine whether the drug being used is a slowly metabolized chemical drug;

[0304] When the node is in use, score item 1 is increased by 1 point; when the node is not in use, it is judged as irrelevant; when the node has R<5.0, SE1-2-1-2 is performed;

[0305] SE1-2-1-2. Determine whether the node with abnormal indicators is "initial treatment" or "subsequent treatment";

[0306] When the node is “initial treatment,” perform SE1-2-1-2-1;

[0307] SE1-2-1-2-1. Determine whether to stop taking the medicine when the indicators begin to become abnormal;

[0308] When the node is not stopped, perform SE1-2-1-2-1-1

[0309] SE1-2-1-2-1-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal;

[0310] When the node is 5-90 days, add 2 points to scoring item 1; when the node is <5 days or >90 days, add 1 point to scoring item 1; when the node is drug withdrawal, perform SE1-2-1-2-1-2;

[0311] SE1-2-1-2-1-2. Calculate the difference between the time of discontinuation of medication and the time when the index begins to become abnormal;

[0312] When the node is ≤30 days, add 1 point to scoring item 1; when the node is >30 days, perform SE1-2-1-2-1-2-1;

[0313] SE1-2-1-2-1-2-1. Determine whether the drug being used is a slowly metabolized chemical drug;

[0314] When the node is used, score item 1 is increased by 1 point; when the node is not used, it is judged as irrelevant; when the node is "subsequent treatment", SE1-2-1-2-2 is performed;

[0315] SE1-2-1-2-2. Determine whether to stop taking the medicine when the indicators begin to become abnormal;

[0316] When the node is not discontinued, perform SE1-2-1-2-2-1;

[0317] SE1-2-1-2-2-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal;

[0318] When the node is 1-90 days, add 2 points to scoring item 1; when the node is >90 days, add 1 point to scoring item 1; when the node is drug withdrawal, perform SE1-2-1-2-2-2;

[0319] SE1-2-1-2-2-2. Calculate the difference between the time of discontinuation of medication and the time when the indicator starts to become abnormal;

[0320] When the node is ≤30 days, add 1 point to scoring item 1; when the node is >30 days, perform SE1-2-1-2-2-2-1;

[0321] SE1-2-1-2-2-2-1. Determine whether the drug being used is a slowly metabolized chemical drug;

[0322] When the node is in use, score item 1 is increased by 1 point; when the node is not in use, it is judged as irrelevant.

[0323] SE2. Get the score of scoring item 2;

[0324] SE2-1. Retrieve medication information and abnormal indicator information to determine whether to stop taking the medication;

[0325] When the node is still in use and there is no conclusion, add 0 points to scoring item 2; when the node is discontinuation, perform SE2-1-1;

[0326] SE2-1-1. Calculate the R value;

[0327] When the node is R ≥ 5.0, perform SE2-1-1-1;

[0328] SE2-1-1-1. Difference between ALT peak and upper limit of normal;

[0329] When the node decreases ≥50% within 8 days, score item 2 adds 3 points; when the node decreases ≥50% within 30 days, score item 2 adds 2 points; when there is no relevant information for the node or it decreases ≥50% after 30 days, score item 2 adds 0 points; when the node decreases <50% after 30 days or increases again, score item 2 adds -2 points; when the node is R<5.0, perform SE2-1-1-2;

[0330] SE2-1-1-2. Difference between peak ALP or TBil value and upper limit of normal;

[0331] When the node decreases by ≥50% within 180 days, 2 points will be added to scoring item 2; when the node decreases by <50% within 180 days, 1 point will be added to scoring item 2; when the node remains unchanged, increases, or there is no data, 0 points will be added to scoring item 2.

[0332] SE3. Get the score of scoring item 3;

[0333] SE3-1. Retrieve the patient's medical history information to determine whether the patient has drinking behavior (current drinking volume: female>2 times / day, male>3 times / day, about 10g ethanol / time);

[0334] When the node is "yes", add 1 point to scoring item 3; when the node is "no", proceed to SE3-2;

[0335] SE3-2. Retrieve patient information and determine whether the patient is pregnant;

[0336] When “No”, add 0 points to scoring item 3; when “Yes”, proceed to SE3-2-1;

[0337] Perform SE3-2-1. Calculate the R value;

[0338] When the node is R<5.0, 1 point is added to scoring item 3; when the node is R≥5.0, 0 point is added to scoring item 3;

[0339] SE3-3. Retrieve the patient's age information and determine whether the patient is over 55 years old;

[0340] When the node is age ≥ 55, score item 3 adds 1 point; when the node is age < 55, score item 3 adds 0 points;

[0341] SE4. Get the score of scoring item 4;

[0342] SE4-1. Retrieve the patient's medication information and determine:

[0343] If there is no or the time of concomitant medication use does not match the time of onset, 0 points will be added to scoring item 4;

[0344] If the duration of concomitant medication use is consistent with the onset of illness, -1 point will be added to item 4;

[0345] If the concomitant medication is known to be hepatotoxic and the duration of use is consistent with the onset of the disease, -2 points will be added to score item 4;

[0346] If there is evidence of liver damage caused by concomitant medication (re-medication reaction or valuable test), score item 4 plus -3 points;

[0347] SE5. Get the score of scoring item 5;

[0348] Other possible causes of liver disease are divided into 2 groups:

[0349] Group I (7 types of causes):

[0350] HAV infection: Anti-HAV-IgM

[0351] HBV infection: HBsAg, anti-HBc-IgM, HBV DNA

[0352] HCV infection: Anti-HCV, HCV RNA

[0353] HEV infection: anti-HEV-IgM, anti-HEV-IgG, HEV RNA

[0354] Hepatobiliary ultrasound imaging / liver vascular color Doppler imaging / endovascular ultrasound examination / CT / MRC

[0355] Alcohol intoxication (AST / ALT ≥ 2)

[0356] Recent history of acute hypotension (especially in the setting of underlying heart disease)

[0357] Group II (5 types of causes):

[0358] Combined with sepsis, metastatic malignant tumors, autoimmune hepatitis, chronic hepatitis B or C, primary biliary cholangitis or primary sclerosing cholangitis, hereditary liver disease, etc.

[0359] CMV infection: anti-CMV-IgM, anti-CMV-IgG, CMV-PCR

[0360] EBV infection: anti-EBV-IgM, anti-EBV-IgG, EBV-PCR

[0361] HSV infection: anti-HSV-IgM, anti-HSV-IgG, HSV-PCR

[0362] VZV infection: anti-VZV-IgM, anti-VZV-IgG, VZV-PCR

[0363] SE5-1. Retrieve the patient's medical history information and determine:

[0364] If all causes in groups I and II can be reasonably excluded, 2 points are added to item 5;

[0365] If all seven causes in group I can be ruled out, but the conditions in group II cannot be ruled out, 1 point is added to item 5;

[0366] If only 5 to 6 causes in group I can be excluded, score item 5 plus 0 points;

[0367] If there are less than 5 causes that can be excluded in group I, 2 points will be subtracted from item 5;

[0368] If the presence of another liver disease is "highly likely", subtract 3 points from item 5;

[0369] SE6. Get the score of scoring item 6;

[0370] SE5-1. Collect reports on previous liver damage caused by drugs and determine:

[0371] If there is a report of hepatotoxicity in the product description, 2 points will be added to the score item 5;

[0372] If there is literature report but no relevant information in the product description, 1 point will be added to the score item 5;

[0373] There were no reports of hepatotoxicity, and the score item 5 plus 0 points;

[0374] SE7. Get the score of scoring item 7;

[0375] SE7-1. Retrieve the patient's medical history information to determine the reaction to medication:

[0376] When the node is R ≥ 5.0,

[0377] If ALT is below 5ULN before medication, and ALT doubles after medication / herbal medicine is used again, 3 points will be added to score item 7;

[0378] If the drug / herbal medicine used in the first reaction is given again and ALT doubles, 1 point is added to score item 7;

[0379] Under the same conditions as the first medication, if ALT is elevated but below ULN, score item 7 is subtracted by 2 points;

[0380] In other cases, 0 points will be added to scoring item 7;

[0381] When the node is R<5.0,

[0382] If the ALP level is below 2ULN before medication, and doubles after medication / herbal medicine is used again, add 3 points to item 7;

[0383] If the drug / herbal medicine used in the first reaction is given again and the ALP level doubles, 1 point will be added to item 7;

[0384] Under the same conditions as the first medication, if ALT is elevated but below ULN, score item 7 is subtracted by 2 points;

[0385] In other cases, add 0 points to scoring item 7.

[0386] 1.2.4. RUCAM scale automatic scoring and DILI analysis algorithm based on explainable artificial intelligence

[0387] Time series analysis, information retrieval and slot grammar are used to automatically score the RUCAM scale and analyze DILI, as shown in Algorithm 1.5. Input: LIS and doctor's advice time series that trigger RUCAM scale scoring, slot grammar, search keywords, medical records, RUCAM scale scoring criteria. Output: RUCAM scale total score, each score and scoring reason.

[0388] Row 1: Based on the time series analysis of the course of disease, score item 1-score the time from taking medicine to onset. Row 2: Based on the time series analysis of the course of disease, score item 2-score the course of disease. Row 3: Based on information retrieval and slot grammar, analyze the course of disease records and score item 3-score the risk factors. Row 4: Based on statistical methods, calculate the number of concomitant medications and score item 4-score the concomitant medications. Row 5: Based on information retrieval and slot grammar, parse the course of disease records and score item 5-score the exclusion of other causes. Row 6: Score item 6-score the previous report of liver injury caused by drugs. Row 7: Based on information retrieval and slot grammar, parse the course of disease records and score item 7-score the response to drug stimulation.

[0389] Line 8: Sum the scores of these seven items and return the total score, the score of each item, and the reason for the score.

[0390]

[0391] 2. Results

[0392] First, we conducted experiments to explore the feasibility and credibility of the explainable artificial intelligence method on three typical cases. Then, we conducted a big data analysis on all 3,896 cases (of which 3,375 cases triggered automatic analysis) to verify the usability of the explainable artificial intelligence method. Finally, we compared the results with the DILI cases manually reported by the hospital to verify the practicality of the explainable artificial intelligence method.

[0393] 2.1 Feasibility and credibility exploration experiments based on typical cases

[0394] For a total of 3896 cases of patients with abnormal liver indicators from December 30, 2018 to September 10, 2023, three typical cases were selected for automatic scoring of the RUCAM scale and DILI analysis.

[0395] 2.1.1. Disease time series experiment

[0396] In Table 1, the time series of the course of disease of three patients are given. Among them, the patient hospitalization ID of the 1st and 2nd columns is J83295 (patient ID is 10053767), the patient hospitalization ID of the 3rd and 4th columns is H61451 (patient ID is 10175795), and the patient hospitalization ID of the 5th and 6th columns is K35375 (patient ID is 10227521). "Suspected drugs" means that the patient has taken suspected drugs that may cause DILI during the period. The quantity can be multiple and the number of times taken can be multiple times. "Normal indicators" means that the liver indicators are normal. "Abnormal indicators" means that the liver indicators are abnormal, which meets the abnormal liver indicator trigger conditions for triggering the automatic scoring of the RUCAM scale and DILI analysis.

[0397] Take the patient with hospitalization ID J83295 as an example. The admission date is 2022-06-16, and the suspected drug was taken on the same day. The suspected drug is at least one of multiple suspected drug conditions. The examination date is 2022-06-17, and the liver index abnormality condition is not met. And during the period (2022-06-17 to 2022-06-25), the suspected drug was taken. The examination date is 2022-06-26, the liver index abnormality condition is not met, and the suspected drug was taken during the period (2022-06-26 to 2022-07-05). The examination date is 2022-07-06, the liver index abnormality condition is not met, and the suspected drug was taken on the same day. The examination date is 2022-07-07, the liver index abnormality condition is not met, and the suspected drug was taken during the period (2022-07-07 to 2022-07-13). The examination date is 2022-07-14, which meets the abnormal liver index conditions, and suspected drugs were taken during the period (2022-07-14 to 2022-07-24). The examination date is 2022-07-25, which does not meet the abnormal liver index conditions, and suspected drugs were taken during the period (2022-07-25 to 2022-08-01). The examination date is 2022-08-02, which does not meet the abnormal liver index conditions, and suspected drugs were taken during the period (2022-08-02 to 2022-08-12). The examination date is 2022-08-13, which does not meet the abnormal liver index conditions, and suspected drugs were taken on the same day.

[0398] 2.1.2. RUCAM automatic scoring and DILI analysis

[0399] The RUCAM automatic scores and DILI analysis of the three patients are given in Table 2. In the RUCAM scale, 0 or less means "excluding" the drug as the cause of liver injury; 1-2 points means "impossible"; 3-5 points means "possible"; 6-8 points means "very likely"; and more than 8 points means "highly likely".

[0400] Table 2 RUCAM automatic score and DILI analysis for patients with hospitalization ID J83295, H61451, and K35375

[0401]

[0402] 2.1.3. Interpretability of RUCAM Scale Automatic Scoring

[0403] 2.1.3.1 Patient 1, whose hospitalization ID is J83695. The total RUCAM score is 5 points. The suspected drug is furosemide tablets. The relevant information is as follows: Time of first abnormal liver index: 2022-07-14 09:00:21. Suspected drug name: furosemide tablets. DILI subtype: cholestatic type. First medication time: 2022-06-16 12:57:27. The AST value of abnormal liver index is: 139.00, and the ALT value is: 164.00. DILI analysis result: possible. The score and interpretation of each item of the RUCAM scale are shown in Table 3.

[0404] Table 3. Automated scoring and interpretation of the RUCAM scale for hospitalization ID J83295

[0405] RUCAM scale scoring items Fraction explain Score 01-Time from medication to onset 2 The number of days from the first medication to the abnormal liver index was 27.84. See Table 1 for the time series of the course of disease. Score 02 - Disease course 0 none Score 03-Risk Factors 1 Patient age: 72 Score 04- Concomitant medication -2 Number of concomitant medications: 17 Score 05-Other reasons for xenophobia 2 none Score 06-Report on previous liver injury caused by medication 2 Previous liver damage report: Yes RUCAM Total Score 5 DILI analysis results possible

[0406] 2.1.3.2 Patient 2, whose hospitalization ID is H61451, has a total RUCAM score of -1. Suspected drug: Oxaliplatin for injection. The relevant information is as follows: Time of first abnormal liver index: 2020-07-31 09:57:13. Suspected drug name: Oxaliplatin for injection. DILI subtype: cholestatic. First medication time: 2020-07-28 09:07:46. The AST value of abnormal liver index is: 210.00, and the ALT value is: 162.00. DILI analysis result: excluded. The score and interpretation of each item of the RUCAM scale are shown in Table 4. In score 05-Exclusion of other reasons, "preliminary diagnosis" and its content come from the semantic analysis of the medical record. The title of the medical record is "Preliminary diagnosis" and the content is "1. Postoperative colon malignant tumor; 2. Postoperative chemotherapy for malignant tumor; 3. Secondary malignant tumor of the liver;". Since there is a liver tumor, 3 points are deducted, excluding other reasons.

[0407] Table 4. Automated scoring and interpretation of the RUCAM scale for hospitalization ID H61451

[0408] RUCAM scale scoring items Fraction explain Score 01-Time from medication to onset 1 The number of days from the first medication to the abnormality of the index: 3.03. See Table 2.1 for the time series of the course of disease. Score 02 - Disease course 0 none Score 03-Risk Factors 1 Patient age: 59 Score 04- Concomitant medication -2 Number of concomitant medications: 2 Score 05-Other reasons for xenophobia -3 Initial diagnosis: 1. Postoperative colon malignancy; 2. Postoperative chemotherapy for malignant tumor; 3. Secondary malignant tumor of the liver; Score 06-Report on previous liver injury caused by medication 2 Previous liver damage report: Yes RUCAM Total Score -1 DILI analysis results exclude

[0409] 2.1.3.3 Patient 3, whose hospitalization ID is K35375. The total RUCAM score is 7 points. The suspected drug is adenosylmethionine butanedisulfonate for injection. The relevant information is as follows: Time of first abnormal liver index: 2022-11-16 09:20:19. Suspected drug: adenosylmethionine butanedisulfonate for injection. DILI subtype: cholestatic. Time of first medication: 2022-11-15 16:41:37. The AST value of abnormal liver index is: 265.00. The ALT value is: 557.00. DILI analysis result: very likely.

[0410] Table 5. Automated scoring and interpretation of the RUCAM scale for patient ID K35375

[0411]

[0412] The score and explanation of each item are shown in Table 5. In score 05-other reasons for exclusion, "discharge diagnosis", "director's first examination_supplementary auxiliary examination" and their contents are derived from the semantic analysis of the medical record. Including the title and content of the medical record. Among them, "high possibility of drug-induced liver damage" has a greater impact on the score. Positive indicators such as "HEV-IgG (+)" are also derived from the medical record.

[0413] In summary, the automatic scoring of the RUCAM scale based on explainable artificial intelligence mainly adopts the time series analysis of the course of disease and the semantic analysis of the course of disease records based on information retrieval and slot grammar. It not only gives the scores of each RUCAM item, but also gives the basis and reasons for the scores. Therefore, this method has strong feasibility and credibility.

[0414] 2.2 Big Data Usability Experiment Based on All Cases

[0415] A total of 3,896 cases of patients with abnormal liver indicators in a tertiary general hospital from December 30, 2018 to September 10, 2023. Some of these patients did not meet the suspected drug trigger conditions or did not take suspected drugs before the liver indicators became abnormal. Therefore, the number of patients who triggered the automatic scoring of the RUCAM scale and DILI analysis was 3,375.

[0416] In the big data experiment, a laptop was used, with each group consisting of 500 cases. The maximum memory occupied by each group analysis was about 300M, and the computing time occupied was about 180 seconds. In other words, using a laptop, the big data analysis can be completed and output to the database in about 30 minutes. This shows that the method based on explainable artificial intelligence for automatic scoring of the RUCAM scale and DILI analysis is available in terms of computing power.

[0417] 2.2.1 Statistical analysis of all cases triggering automatic RUCAM scoring and DILI analysis

[0418] 2.2.1.1 Overview Number of times the RUCAM scale was triggered for automatic scoring: 24,160 times. A patient may take multiple suspected drugs and trigger the RUCAM scale automatically multiple times. Number of patient IDs: 3,375 cases. Number of hospitalization IDs: 3,784 cases. Some patients may have been hospitalized multiple times and have multiple hospitalization IDs. 1) Distribution of DILI subtypes. Hepatocellular: 2,979 times, mixed: 6,241 times, cholestatic: 14,940 times. 2) Gender distribution. Male: 14,935 times, female: 9,225 times.

[0419] 2.2.1.2 Distribution of total scores of all cases triggering RUCAM scale automatic scoring and DILI analysis As can be seen from Table 6, among all cases triggered by RUCAM scale automatic scoring and DILI analysis, the proportion of "exclusion" was the highest, accounting for about 47.13%. The number of "highly likely" cases was the least, with only 1 case. The proportion of "very likely" was 9.70%, and patients with "highly likely" and "very likely" were the focus of DILI analysis. The combined proportion of these two groups of patients was less than 10%, which greatly reduced the burden of manual reporting of ADRs.

[0420] Table 6. Distribution of total scores for all cases triggering automatic RUCAM scoring and DILI analysis

[0421]

[0422] 2.2.1.3 The age distribution of patients that triggered the automatic scoring of the RUCAM scale and DILI analysis is shown in Table 7. The top three age distributions of patients are: 0-9 years old, 60-69 years old, and 50-59 years old, which may be related to the age distribution of inpatients in the hospital. There are a large number of pediatric patients in a tertiary general hospital.

[0423] Table 7. Age distribution of patients triggering RUCAM automatic scoring and DILI analysis

[0424] Age group (years) 0-9 10-19 20-29 30-39 40-49 50-59 60-69 70-79 80-89 ≥90 Trigger times 5661 1363 478 1127 1694 3728 5052 3414 1278 365 Trigger times ratio (%) 23.43 5.64 1.98 4.66 7.01 15.43 20.91 14.13 5.29 6.80

[0425] 2.2.1.4 Distribution of drug names and triggering times that trigger RUCAM scale automatic scoring and DILI analysis The top 10 drugs with the most triggering times were selected, including drug name and triggering times (values ​​in brackets). The order from high to low is: dexamethasone sodium phosphate injection (1148), furosemide injection (993), propofol medium / long chain fat emulsion injection (574), dexmedetomidine hydrochloride injection (548), omeprazole sodium for injection (545), cefuroxime sodium for injection (512), cefoperazone sodium and sulbactam sodium for injection (506), penicillin sodium for injection (470), bromhexine hydrochloride for injection (442), and hydromorphone hydrochloride injection (408).

[0426] 2.2.2 Statistical analysis of “probable and highly probable” cases triggering automatic RUCAM scoring and DILI analysis

[0427] 2.2.2.1 Overview The number of times the RUCAM scale was triggered with an automatic score greater than or equal to 6 points ("very likely" and "highly likely"): 2344. Number of patient IDs: 481. Number of hospitalized IDs: 490. Among them, 1) Distribution of DILI subtypes. Hepatocellular: 551 times, mixed: 365 times, cholestatic: 1428 times. It can be seen that the cholestatic type was triggered the most times. 2) Gender distribution. Male: 1507 cases, female: 837 cases. It can be seen that the number of male patients is more than that of female patients.

[0428] 2.2.2.2 Distribution of “very likely and highly likely” cases. Very likely: 2343 times. Highly likely: 1 time. It can be seen that there are very few “highly likely” cases.

[0429] 2.2.2.3 Age distribution of “probable and highly probable” cases As shown in Table 8, the top three age distribution groups of patients in “probable and highly probable” cases are: 60-69 years old, 0-9 years old, and 50-59 years old.

[0430] Table 8. Age distribution of patients triggering RUCAM scale automatic scoring and DILI analysis for “probable and highly probable” cases

[0431] Age group (years) 0-9 10-19 20-29 30-39 40-49 50-59 60-69 70-79 80-89 ≥90 Trigger times 457 104 58 74 133 396 556 368 113 84 Trigger times ratio (%) 19.50 4.43 2.47 3.16 5.67 16.89 23.72 15.70 4.82 3.58

[0432] 2.2.2.4 Distribution of drugs with the most “very likely and highly likely” cases The top 10 drugs with the most triggering times are selected, including drug name and triggering times (values ​​in brackets). The order from high to low is: dexamethasone sodium phosphate injection (113), furosemide injection (109), dexmedetomidine hydrochloride injection (81), propofol medium / long chain emulsion injection (62), omeprazole sodium for injection (60), penicillin sodium for injection (60), cefuroxime sodium for injection (56), cefoperazone sodium and sulbactam sodium for injection (55), vancomycin hydrochloride for injection (54), and nifedipine tablets (47).

[0433] 2.2.3 Analysis of typical cases triggering automatic scoring of RUCAM scale and DILI analysis

[0434] 2.2.3.1 DILI subtype is hepatocellular type. Take patient ID 74383692, hospitalization ID H50926, and suspected drug vancomycin hydrochloride for injection as an example. Time of first abnormal liver index: 2020-07-03 09:32:15. DILI subtype: hepatocellular type. Time of first medication: 2020-07-01 08:06:14. The value of AST for abnormal liver index is: 684.00. The value of ALT is: 744.00. The score and interpretation of each item of the RUCAM scale are shown in Table 2.9. In score 02-course, the subsequent difference is 608.00, the peak difference is 624.00, the ratio of the two is greater than 0.5, and the number of days of drug withdrawal is 3.91 (within 8 days), so the score of score 02-course is 3.0.

[0435] Table 9. Results and interpretation of the RUCAM automatic scoring scale for hospitalization ID H50926

[0436]

[0437] 2.2.3.2 The patient ID of the “highly likely” DILI case is 70966632 and the hospitalization ID is K37593. Time of first abnormal liver index: 2022-11-22 09:25:43. Suspected drug name: Losartan potassium hydrochlorothiazide tablets. DILI subtype: hepatocellular type. First medication time: 2022-11-21 15:05:28. The AST value of abnormal liver index is: 307.00, and the ALT value is: 685.00. The score and interpretation of each item of the RUCAM scale are shown in Table 10. In the score 05-excluding other reasons, the title and content of the medical record are: “Current diagnosis: 1. Type 2 diabetes with poor blood sugar control;

[0438] 2. Type 2 diabetic peripheral neuropathy; 3. Type 2 diabetic peripheral vascular disease; 4. Hypertension stage 2; 5. Personal history of endometrial malignancy; 6. Lung shadows; 7. Liver damage; 8. Renal damage. Since "liver damage" is included in the diagnosis, 2 points are added to this score.

[0439] Table 10. Results and interpretation of the RUCAM automatic scoring scale for hospitalization ID H50926

[0440]

[0441] In summary, the automatic scoring of the RUCAM scale and DILI analysis based on explainable artificial intelligence was performed on all 3896 cases, with 500 cases in each group, which took about 30 minutes on a laptop computer, indicating that this method has good usability.

[0442] 2.3 Practical experiment of successful reporting of DILI cases by hospitals

[0443] 2.3.1. Hospital successfully reported DILI cases

[0444] Nine cases of successful DILI reported by inpatients in a tertiary general hospital from June 2023 to May 2024 were selected for analysis. The patient numbers are {1, 2, 3, 4, 5, 6, 7, 8, 9}, as shown in Table 11. Including patient ID, hospitalization ID, suspected drugs, reporting results, etc. After preliminary computer analysis, these cases were divided into invalid cases and valid cases.

[0445] Table 11. Some cases of DILI reported successfully by the hospital

[0446]

[0447]

[0448] 2.3.2. Invalid Case

[0449] Invalid cases refer to cases that did not trigger the automatic scoring and DILI analysis of the RUCAM scale, including four patients with patient numbers {1, 2, 4, 8}. As mentioned above, the prerequisite for triggering computer DILI analysis is abnormal liver indicators and suspected drugs have been taken before the liver indicators became abnormal. Among them, patient {1} was found to have abnormal liver indicators, but did not take suspected drugs before the liver indicators became abnormal (suspected drugs were taken after the liver indicators became abnormal). Patients {2, 4, 8} did not have abnormal liver indicators (only hypothermia was found). Since these four patients did not meet the triggering conditions for the automatic scoring and DILI analysis of the RUCAM scale, they were judged to be invalid cases. As listed in Table 2.

[0450] Table 12. Invalid cases that did not trigger RUCAM scale automatic scoring and DILI analysis

[0451]

[0452] 2.3.3. Effective cases

[0453] Valid cases refer to those that meet the RUCAM scale automatic scoring and DILI analysis trigger conditions, and the patient numbers are {3, 5, 6, 7, 9}.

[0454] 2.3.3.1 Comparison of manual reporting and computer analysis results of valid cases Among the valid cases, the manually reported results of 3 cases {3, 6, 9} were the same as the results of computer analysis. The manually reported hospitalization ID of case {7} was K44256 (result "very likely"), which did not trigger computer analysis; computer analysis was triggered when the hospitalization ID was K48841 in another hospitalization (result "possible"). See Table 13 for computer analysis results, RUCAM scores, and DILI subtypes. The computer analysis results and manually reported results were the same in 3 cases and similar in 2 cases. The practicality of the automatic scoring and DILI analysis method based on the RUCAM scale was demonstrated.

[0455] Table 13. Table of valid cases for RUCAM automatic scoring and DILI analysis. Including comparison of valid cases reported manually and analyzed by computer, RUCAM score, DILI subtype. Among them, the hospitalization ID of patient {7} reported manually was K44256 (no computer analysis was triggered), and the hospitalization ID that triggered computer analysis was K48841

[0456]

[0457] 2.3.3.2 Complete results of computer analysis of valid cases For the successfully reported patient hospitalization ID, the DILI results of other suspected drugs during this hospitalization period in addition to the reported drugs were analyzed, as shown in Table 14. In addition to the drugs reported by the hospital, 17 suspected drugs (cases) were found, and their RUCAM scores were all higher than 4 points (possible), marked in blue in the table. Note: There are no manually reported results for these suspected drugs (cases). As can be seen from Table 13, the results of automatic scoring and DILI analysis based on the RUCAN scale are more than the suspected cases found manually. In this experiment, 5 suspected cases were found manually, and the computer found another 17 suspected cases in these hospitalization IDs, which is about 3 times the number of manual analysis.

[0458] Table 14. Complete results of RUCAM scale automatic scoring and DILI analysis for effective cases. All suspected drugs were analyzed based on the original manually reported suspected drugs, and 17 new suspected drugs or effective cases were added, which are shown in bold in the table.

[0459]

[0460] In summary, among the 9 suspected cases reported manually, 4 did not trigger the RUCAM scale automatic scoring and DILI analysis, and 5 triggered it. Among the 5 triggered cases, the results of computer analysis were the same as those of manual analysis in 3 cases, and similar in 2 cases. Therefore, as long as the RUCAM scale automatic scoring and DILI analysis can be triggered, the accuracy of the explainable artificial intelligence method is close to that of manual analysis, and its coverage or underreporting rate is much better than that of manual analysis.

[0461] 3 Conclusion

[0462] This embodiment performs automatic RUCAM scale scoring and DILI analysis on the big data of abnormal liver indicators in a tertiary comprehensive hospital (3896 cases) and some successfully reported DILI cases (9 cases). It mainly adopts explainable artificial intelligence methods, including time series analysis of the course of disease and semantic analysis of medical records based on information retrieval and slot grammar. In the exploratory experiment of typical cases, the feasibility and usability of the explainable artificial intelligence method were verified. In the big data experiment of 3375 cases that triggered the automatic scoring of the RUCAM scale, a total of 14160 automatic scorings were triggered, and the total score of the RUCAM scale, the scores of each item and the basis for scoring, the distribution of DILI subtypes, patient age, and patient gender, as well as the distribution of suspected drugs with the most triggering times were given, and typical cases were analyzed. It takes about 30 minutes to analyze on a laptop, and the maximum memory occupied is about 300M, which verifies the usability of this method. Among the 9 DILI cases successfully reported by the hospital, 5 triggered the automatic scoring of the RUCAM scale and DILI analysis. The analysis results were the same (3 cases) or similar (2 cases) to those reported manually. In addition, the results of computer analysis found more suspected cases, verifying the practicality of this method.

[0463] Although the above description is centered on the embodiment, it is only an example and does not limit the present invention. It is clear to those skilled in the art that various modifications and applications not illustrated above can be made within the scope of the essential characteristics of the present embodiment. For example, each component specifically shown in the embodiment can be implemented after being modified. Moreover, various differences related to such modifications and applications should be interpreted as being included in the scope of the present invention as defined in the attached claims.

Claims

1. RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence, characterized by: The time series method is used to analyze the sequence between patients' medical orders and LIS indicators. Information retrieval and slot grammar are used to perform semantic analysis on patients' medical records, thereby realizing automatic scoring of the RUCAM scale and DILI analysis, and providing an explainable scoring basis.

2. The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence according to claim 1, characterized in that: The following steps are included: S1. Retrieve the patient's medical information and determine whether the patient's medical information contains conditions that trigger the automatic scoring of the RUCAM scale. When "included", proceed to S2; S2. Scoring is performed based on the RUCAM scale items, and the DILI analysis results and basis are given; Among them, the disease course time series needs to be constructed in the process of determining whether the patient's diagnosis and treatment information contains the conditions that trigger the automatic scoring of the RUCAM scale; The construction of the disease course time series is a disease course time series analysis for automatic scoring of the RUCAM scale, which requires obtaining the LIS indicators and medical advice along the timeline, and then analyzing the time sequence, calculating the length of the time interval and information related to the timeline, and constructing the disease course time series based on this.

3. The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence as claimed in claim 2, characterized in that: The conditions that trigger the automatic scoring of the RUCAM scale include: abnormal liver index conditions and suspected drug conditions.

4. The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence as claimed in claim 2, characterized in that: The method for constructing the disease course time series is as follows: SA1. Extract LIS indicators; SA2. Sort by time; SA3. Determine whether there is an abnormality in the LIS item, and mark it if there is an abnormality; SA4. Determine whether there is an abnormality in the liver index LIS group, and mark it if there is an abnormality; SA5. Extract the doctor's orders and determine whether there are any suspected drugs. If there are any suspected drugs, mark them; SA6. Combine LIS indicators and physician orders; SA7. Sort LIS indicators and physician orders by time.

5. The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence according to claim 4, characterized in that: The method for determining whether the patient's diagnosis and treatment information contains conditions for triggering automatic scoring of the RUCAM scale is as follows: SC1. Based on the time series of the course of disease, determine whether there are abnormal markers of the liver index LIS group; When present, SC2 was performed; SC2. Determine whether there are suspected drug markers from the date of admission to the date of abnormal liver indicators; When present, it triggers the condition for automatic scoring of the RUCAM scale.

6. The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence according to claim 1, characterized in that: The scoring based on the RUCAM scale items includes scoring item 1 based on the time series analysis from medication to onset, scoring item 2 based on the analysis of the course of the disease, scoring item 3 based on risk factors, scoring item 4 based on the evaluation of concurrently used medications, scoring item 5 based on the exclusion of other reasons, scoring item 6 based on previous reports of liver damage caused by drugs, and scoring item 7 based on the response to drug re-stimulation.

7. The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence according to claim 6, characterized in that: Before scoring each scoring item, DILI typing is also required; The method of DILI typing is as follows: SD1. Retrieve ALT and ALP data; SD2. Get the R value based on R = [ALT / ALT upper limit of normal range] / [ALP / ALP upper limit of normal range]; SD3. When R ≥ 5.0, it is marked as hepatocellular type; When R≦2.0, it is marked as cholestatic type; Others, labeled as mixed.

8. The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence according to claim 7, characterized in that: The specific method for scoring based on the RUCAM scale items is as follows: SE1. Get the score of scoring item 1; SE1-1. Retrieve medication information and indicator abnormality information, and determine whether the indicator abnormality node is before "starting medication" or after "starting medication"; When the node is before "start taking medicine", it is judged as irrelevant; When the node is "start taking medication", perform SE1-2; SE1-2. Get the time of first taking the medicine and the time when the abnormality of the index began. When the node is such as the time from medication to onset cannot be obtained, it is considered as unavailable for evaluation; When the node is the time from medication availability to onset, SE1-2-1; SE1-2-1. Calculate the R value; When the node is R ≥ 5.0, perform SE1-2-1-1; SE1-2-1-1. Determine whether the node with abnormal indicators is "initial treatment" or "subsequent treatment"; When the node is "initial treatment", perform SE1-2-1-1-1; SE1-2-1-1-1. Determine whether to stop taking the medicine when the indicators begin to become abnormal; When the node is not stopped, perform SE1-2-1-1-1-1 SE1-2-1-1-1-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal; When the node is 5-90 days, score item 1 will be increased by 2 points; When the node is <5 days or >90 days, score item 1 is increased by 1 point; When the node is in drug withdrawal, perform SE1-2-1-1-1-2; SE1-2-1-1-1-2. Calculate the difference between the time of discontinuation of medication and the time when the indicator starts to become abnormal; When the node is ≤15 days, score item 1 is increased by 1 point; When the node is >15 days, perform SE1-2-1-1-1-2-1; SE1-2-1-1-1-2-1. Determine whether the drug being used is a slowly metabolized chemical drug; When the node is in use, score item 1 is increased by 1 point; When a node is unused, it is judged as irrelevant; When the node is "Subsequent Treatment", perform SE1-2-1-1-2; SE1-2-1-1-2. Determine whether to stop taking the medicine when the indicators start to become abnormal; When the node is not discontinued, perform SE1-2-1-1-2-1; SE1-2-1-1-2-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal; When the node is 1-15 days, score item 1 will be increased by 2 points; When the node is >15 days, score item 1 is increased by 1 point; When the node is in drug withdrawal, perform SE1-2-1-1-2-2; SE1-2-1-1-2-2. Calculate the difference between the time of discontinuation of medication and the time when the indicator starts to become abnormal; When the node is ≤15 days, score item 1 is increased by 1 point; When the node is >15 days, perform SE1-2-1-1-2-2-1; SE1-2-1-1-2-2-1. Determine whether the drug being used is a slowly metabolized chemical drug; When the node is in use, score item 1 is increased by 1 point; When a node is unused, it is judged as irrelevant; When the node is R<5.0, perform SE1-2-1-2; SE1-2-1-2. Determine whether the node with abnormal indicators is "initial treatment" or "subsequent treatment"; When the node is "initial treatment", perform SE1-2-1-2-1; SE1-2-1-2-1. Determine whether to stop taking the medicine when the indicators start to become abnormal; When the node is not stopped, perform SE1-2-1-2-1-1 SE1-2-1-2-1-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal; When the node is 5-90 days, score item 1 will be increased by 2 points; When the node is <5 days or >90 days, score item 1 is increased by 1 point; When the node is in drug withdrawal, perform SE1-2-1-2-1-2; SE1-2-1-2-1-2. Calculate the difference between the time of discontinuation of medication and the time when the indicator starts to become abnormal; When the node is ≤30 days, score item 1 is increased by 1 point; When the node is >30 days, perform SE1-2-1-2-1-2-1; SE1-2-1-2-1-2-1. Determine whether the drug being used is a slowly metabolized chemical drug; When the node is in use, score item 1 is increased by 1 point; When a node is unused, it is judged as irrelevant; When the node is "Subsequent Treatment", perform SE1-2-1-2-2; SE1-2-1-2-2. Determine whether to stop taking the medicine when the indicators begin to become abnormal; When the node is not discontinued, perform SE1-2-1-2-2-1; SE1-2-1-2-2-1. Calculate the difference between the time of first medication and the time when the index begins to become abnormal; When the node is 1-90 days, score item 1 plus 2 points; When the node is >90 days, score item 1 is increased by 1 point; When the node is in drug withdrawal, perform SE1-2-1-2-2-2; SE1-2-1-2-2-2. Calculate the difference between the time of discontinuation of medication and the time when the indicator starts to become abnormal; When the node is ≤30 days, score item 1 is increased by 1 point; When the node is >30 days, perform SE1-2-1-2-2-2-1; SE1-2-1-2-2-2-1. Determine whether the drug being used is a slowly metabolized chemical drug; When the node is in use, score item 1 is increased by 1 point; When a node is unused, it is judged as irrelevant; SE2. Get the score of scoring item 2; SE2-1. Retrieve medication information and abnormal indicator information to determine whether to stop taking the medication; When the node is still in use and there is no conclusion, score item 2 is increased by 0 points; When the node is drug withdrawal, perform SE2-1-1; SE2-1-1. Calculate the R value; When the node is R ≥ 5.0, perform SE2-1-1-1; SE2-1-1-1. Difference between ALT peak and upper limit of normal; If the node drops by ≥50% within 8 days, 3 points will be added to scoring item 2; If the node is reduced by ≥50% within 30 days, 2 points will be added to scoring item 2; If there is no relevant data for a node or it decreases by ≥50% after 30 days, score item 2 will be increased by 0 points; If the node decreases by <50% or increases again after 30 days, score item 2 will be increased by -2 points; When the node is R<5.0, perform SE2-1-1-2; SE2-1-1-2. Difference between peak ALP or TBil value and upper limit of normal; If the node drops by ≥50% within 180 days, 2 points will be added to scoring item 2; If the node drops by less than 50% within 180 days, score item 2 will be increased by 1 point; When the node is unchanged, rising, or has no data, score item 2 is increased by 0 points. SE3. Get the score of scoring item 3; SE3-1. Retrieve the patient's medical history information to determine whether the patient has drinking behavior (current drinking volume: female>2 times / day, male>3 times / day, about 10g ethanol / time); When the node is "yes", score item 3 is increased by 1 point; When the node is "No", proceed to SE3-2; SE3-2. Retrieve patient information and determine whether the patient is pregnant; When "No", score item 3 will be added with 0 points; If "yes", proceed to SE3-2-1; Perform SE3-2-1. Calculate the R value; When the node is R<5.0, score item 3 is increased by 1 point; When the node is R≥5.0, score item 3 is increased by 0 points; SE3-3. Retrieve the patient's age information and determine whether the patient is over 55 years old; When the node is age ≥ 55, score item 3 is increased by 1 point; When the node is age < 55, score item 3 is added with 0 points; SE4. Get the score of scoring item 4; SE4-1. Retrieve the patient's medication information and determine: If there is no or the time of concomitant medication use does not match the time of onset, 0 points will be added to scoring item 4; If the duration of concomitant medication use is consistent with the onset of illness, -1 point will be added to item 4; If the concomitant medication is known to be hepatotoxic and the duration of use is consistent with the onset of the disease, -2 points will be added to score item 4; If there is evidence of liver damage caused by concomitant medication (re-medication reaction or valuable test), score item 4 plus -3 points; SE5. Get the score of scoring item 5; Other possible causes of liver disease are divided into 2 groups: Group I: HAV infection: Anti-HAV-IgM HBV infection: HBsAg, anti-HBc-IgM, HBV DNA HCV infection: Anti-HCV, HCV RNA HEV infection: anti-HEV-IgM, anti-HEV-IgG, HEV RNA Hepatobiliary ultrasound imaging / liver vascular color Doppler imaging / endovascular ultrasound examination / CT / MRC Alcoholism Recent history of acute hypotension Group II: Patients with concurrent sepsis, metastatic malignant tumors, autoimmune hepatitis, chronic hepatitis B or C, primary biliary cholangitis or primary sclerosing cholangitis, or hereditary liver disease; CMV infection: anti-CMV-IgM, anti-CMV-IgG, CMV-PCR EBV infection: anti-EBV-IgM, anti-EBV-IgG, EBV-PCR HSV infection: anti-HSV-IgM, anti-HSV-IgG, HSV-PCR VZV infection: anti-VZV-IgM, anti-VZV-IgG, VZV-PCR SE5-1. Retrieve the patient's medical history information and determine: If all causes in groups I and II can be reasonably excluded, 2 points are added to item 5; If all seven causes in group I can be ruled out, but the conditions in group II cannot be ruled out, 1 point is added to item 5; If only 5 to 6 causes in group I can be excluded, score item 5 plus 0 points; If there are less than 5 causes that can be excluded in group I, 2 points will be subtracted from item 5; If the presence of another liver disease is "highly likely", subtract 3 points from item 5; SE6. Get the score of scoring item 6; SE6-1. Collect reports on previous liver damage caused by drugs and determine: If there is a report of hepatotoxicity in the product description, 2 points will be added to the score item 5; If there is literature report but no relevant information in the product description, 1 point will be added to the score item 5; There were no reports of hepatotoxicity, and the score item 5 plus 0 points; SE7. Get the score of scoring item 7; SE7-1. Retrieve the patient's medical history information and determine the reaction to medication: When the node is R ≥ 5.0, If ALT is below 5ULN before medication, and ALT doubles after medication / herbal medicine is used again, 3 points will be added to score item 7; If the drug / herbal medicine used in the first reaction is given again and ALT doubles, 1 point is added to score item 7; Under the same conditions as the first medication, if ALT is elevated but below ULN, score item 7 is subtracted by 2 points; In other cases, 0 points will be added to scoring item 7; When the node is R<5.0, If the ALP level is below 2ULN before medication, and doubles after medication / herbal medicine is used again, add 3 points to item 7; If the drug / herbal medicine used in the first reaction is given again and the ALP level doubles, 1 point is added to score item 7; Under the same conditions as the first medication, if ALT is elevated but below ULN, score item 7 is subtracted by 2 points; In other cases, add 0 points to scoring item 7.

9. The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence according to claim 2, characterized in that: Analytical method of the DILI analysis: When the RUCAM scale is scored, 0 or less means "excluding" the drug as the cause of liver injury; When the RUCAM scale score is 1 to 2 points, it means "impossible"; When the RUCAM scale score is 3 to 5 points, it means "possible"; When the RUCAM scale score is 6 to 8 points, it means "very likely"; A RUCAM score greater than 8 indicates "high probability".

10. The RUCAM scale automatic scoring and DILI analysis method based on explainable artificial intelligence according to claim 1, characterized in that: The method for semantically analyzing patient medical records using information retrieval and slot grammar is as follows: SE1. To meet the needs of RUCAM scale scoring, information retrieval was performed in the content field based on the search terms; If "existent", proceed to SE2; If "does not exist", repeat SE1; SE2. Determine whether the search term is in the title; When "yes", filter and repeat SE1; When "No", proceed to SE3; SE3. Perform sentence division and slot grammar parsing.

Images