A method of quantifying case complexity and acuity

Abstract

The application provides a method for quantifying case complexity and difficulty, comprising the following steps: collecting sample data, preprocessing the sample data to extract valid medical record data; performing scale processing on the medical record data to identify disease patterns and trends in the medical record data; inputting the medical record data into a DRG grouper based on the disease patterns and trends for grouping calculation, with each case entering a DRG group; obtaining a weight parameter of the case and performing standardization processing on the parameter to calculate the case complexity and difficulty. By constructing a medical record prediction model and using machine learning technology to identify disease patterns and trends, the application can predict the severity of the disease or the treatment result according to the existing characteristics of the patient, assist in medical decision-making, and improve the treatment effect and the recovery rate of the patient. Through analysis of large-scale medical record data, the application assists in studying disease patterns, correlation and development trends, and improves user experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical informatization, in particular to a method for quantifying case complexity and difficulty level. BACKGROUND

[0002] The workload performance evaluation reform of hospitals requires changing the original quantity (person-time) evaluation to quality (case complexity and difficulty level) evaluation, and the case complexity and difficulty level need to be quantified to facilitate performance evaluation. The existing technologies all use fees to measure the complexity and difficulty level of cases, but the treatment fees of cases are composed of bed fees, examination fees, laboratory fees, drug fees, consumable fees, treatment fees, surgery fees, nursing fees, etc. Due to the current situation of the country, pricing does not fully reflect the technical value of doctors and nursing staff, and the same case has a huge impact on fees such as drug fees and consumable fees. The same effect, different manufacturers of drugs and consumables have a large difference in fees, and it is impossible to quantify the complexity and difficulty level of cases by fees.

[0003] DRGs (Diagnosis Related Groups) is a Chinese term for (disease) diagnosis related grouping. According to the patient's age, gender, clinical diagnosis, complications and comorbidities, surgery, operation, outcome, length of stay, and other factors, the patient is classified into related groups. Its definition includes three parts:

[0004] First, it is a patient classification scheme. The core idea of DRGs is to group cases with the same characteristics in one group to facilitate management.

[0005] Second, the basis of DRGs classification is the patient's diagnosis and surgery, and the influence of the patient's age, complications and comorbidities is considered.

[0006] Third, it links the hospital's treatment of patients and the costs incurred, thereby providing a basis for the development of payment standards, especially the implementation of prepayment.

[0007] DRGs technology first classifies cases, groups them by DRGs, and cases in the same DRG group are similar in disease, treatment method, and comorbidities / complications, and should have the same complexity and difficulty level. For each DRG group, a weight value is determined to express the complexity and difficulty level of the disease group, expressing the complexity and difficulty level of the cases in the disease group.

[0008] The existing DRG grouper is applied to medical insurance cost payment, which has certain difference from clinic, mainly considering that patients with the same disease and the same treatment method are divided into the same DRG group. If the cost difference in the group is large, the group can be split into multiple subgroups according to the number of hospitalization days to achieve the purpose of small cost difference in the DRG group, so as to be applied to medical insurance payment. The payment DRG grouper mainly considering cost is not suitable for performance application, which does not conform to the clinical reality. At the same time, using the average total cost or sub-item cost weighting method as the weight (CW) only reflects the cost difference. For the same disease and the same diagnosis and treatment process, the same efficacy and different drug and consumable products are adopted, and the cost difference is huge. However, for clinicians, the disease severity, treatment complexity and medical resource consumption are the same. CW cannot completely reflect the criticality of the case, the complexity of the diagnosis and treatment process and the consumption of medical resources. Only considering cost is not enough, and multiple parameters need to be considered to conform to the clinical reality.

[0009] Therefore, it is desirable to have a method for quantifying case complexity and difficulty based on DRGs to solve the problems in the prior art. SUMMARY

[0010] To solve the above problems, the present application provides a method for quantifying case complexity and difficulty, which is based on disease diagnosis related grouping to calculate case complexity and difficulty by using relative weight. The method comprises the following steps:

[0011] S10: Collecting sample data, pre-processing the sample data to extract valid medical record data;

[0012] S20: Scaling the medical record data to identify disease patterns and trends in the medical record data;

[0013] S30: Based on the disease patterns and trends, inputting the medical record data into a DRG grouper for grouping calculation, and each case entering a DRG group;

[0014] S40: Obtaining the weight parameters of the case and standardizing the parameters to calculate the case complexity and difficulty.

[0015] Optionally, the step of collecting sample data and pre-processing the sample data to extract valid medical record data comprises at least one of the following:

[0016] Performing a data cleaning process to remove duplicate data in the sample data, processing missing values in the sample data, and checking logical errors and inconsistencies in the data, and correcting error data;

[0017] Data conversion is performed on data in different data formats, numerical sample data of different dimensions and ranges are converted into a unified standard scale, continuous data and analog data are converted into discrete data according to the requirements of quantization processing, and classification data in medical records are converted into digital coding forms;

[0018] Data integration processing is performed, the medical record data of different sources are loaded into a unified database for integration, and the data in different systems are associated according to the identity of the patient;

[0019] The original data source is called to verify the accuracy of the sample data, and the consistency of the medical record data in different records is checked.

[0020] Optionally, the process of performing large-scale processing on the medical record data includes:

[0021] Learning feature variables related to a specific disease from the medical record data using natural language processing technology;

[0022] Selecting and constructing feature variables related to disease diagnosis and treatment effect, aggregating similar cases into multiple class clusters, and mining the association between feature variables in medical record data;

[0023] Building a medical record prediction model to identify disease patterns and trends in the medical record data using a machine learning model.

[0024] Optionally, the process of learning feature variables related to a specific disease from the medical record data using natural language processing technology includes:

[0025] Segmenting text data in the medical record data into individual words and annotating the part of speech for each word;

[0026] Analyzing the syntax structure of each sentence, determining the dependency relationship between words, constructing a syntax tree, and extracting key phrases related to diseases according to the syntax structure;

[0027] Identifying medical entities mentioned in the text data in the medical record data, and classifying and annotating the identified medical entities;

[0028] According to the medical characteristics, analyze the classified and annotated medical entities, and according to the preset statistical method, calculate the importance of each word in the medical record data to extract feature variables;

[0029] Converting the extracted feature variables into vector representation using word embedding technology to perform importance analysis on the feature variables.

[0030] Optionally, the word embedding technique includes the following in converting the extracted feature variables into vector representations:

[0031] A vocabulary table is constructed based on the feature variables of the medical record data, and each word in the vocabulary table is converted into a fixed-length vector using a pre-trained word embedding model;

[0032] According to the converted vector of each word, the medical record is represented as a vector sequence to input the machine learning model for training or prediction.

[0033] Optionally, the steps of identifying disease patterns and trends in the medical record data using a machine learning model and before include:

[0034] Based on the multiple clusters and association relationships, a comprehensive application is made based on Naive Bayes, Support Vector Machine, Random Forest, and Neural Network to conditionally judge the features in the medical record data and divide the medical record data into different disease categories;

[0035] A regression model is established to predict the relationship between the hospitalization days and the age of patients corresponding to different diseases.

[0036] Optionally, the process of dividing the medical record data into different disease categories based on the multiple clusters and association relationships, and comprehensively applying Naive Bayes, Support Vector Machine, Random Forest, and Neural Network to conditionally judge the features in the medical record data includes:

[0037] A model fusion technique is used to combine the prediction results of multiple models in a stacking manner to determine the final classification result.

[0038] Optionally, the process of using a model fusion technique to combine the prediction results of multiple models in a stacking manner to determine the final classification result includes:

[0039] Based on the features of the medical record data, Naive Bayes is used to preliminarily classify the medical record data to obtain a first classification result;

[0040] The first classification result is used as the input feature of the Support Vector Machine, and other features of the medical record data are combined to perform a second classification on the medical record data to obtain a second classification result;

[0041] The first classification result, the second classification result, and other features in the medical record data are used as the input of part of the decision trees in the Random Forest to perform a third classification on the medical record data to obtain a third classification result.

[0042] Optionally, the process of inputting the medical record data into the DRG grouper for grouping calculation based on the disease mode and trend includes:

[0043] Grouping according to the MDC major category based on the data of the medical record header;

[0044] Dividing each MDC into a disease category, using the MDC A preliminary grouping disease;

[0045] Subdividing each disease category into DRGs groups according to age, gender and severity of complications;

[0046] Verifying each DRGs group using big data, and adjusting the group according to the CV coefficient of variation;

[0047] Grouping the cases according to the parameters of age, gender, neonatal admission weight, neonatal discharge weight, marriage, admission and discharge date, main diagnosis, secondary diagnosis and surgery in the medical record data.

[0048] Optionally, the process of obtaining the weight parameters of the cases and standardizing the parameters to calculate the complexity and difficulty of the cases includes:

[0049] The weight parameters include: primary care proportion pA, mortality rate pB, average hospitalization days pC and cost pD, and the standardization calculation formula of each parameter is formula (1)-(4):

[0050] Primary care proportion pA = corresponding DRG grouping primary care proportion value / total sample primary care proportion (1);

[0051] Mortality rate pB = number of deaths in a certain DRG group / total number of people in the DRG group (2);

[0052] Average hospitalization days pC = average hospitalization days of the corresponding DRG group / average hospitalization days of the total sample (3);

[0053] Cost pD = average medical cost of the corresponding DRG group / average medical cost of the total sample (4);

[0054] The weight value RW of the complexity and difficulty of the cases is calculated according to the weight parameters, and the weight value RW calculation formula (5) is:

[0055] RW = pA*r1+pB*r2+pC*r3+pD*r4 (5)

[0056] Wherein, r1+r2+r3+r4 = 1, r1 represents the severity of the disease; r2 represents the risk degree of the disease; r3 represents the degree of human resource consumption; r4 represents the degree of medical resource consumption.

[0057] The application develops a performance DRG grouper from the disease types in the clinical front line. The performance weight of the disease group is calculated from the criticality, complexity, occupied human resources and consumed medical resources of the disease. The performance weight RW (Relative Weight) of the disease group is calculated by using the weighted calculation, which solves the weight value quantization problem close to the clinic, reflecting the complexity and difficulty of the case. The application can predict the severity of the disease or the treatment result according to the existing characteristics of the patient by constructing a medical record prediction model and using machine learning technology to identify disease patterns and trends, which can provide a reference for doctors to develop personalized treatment plans, assist medical decision-making, improve treatment effect and patient recovery rate. The technical scheme of the application provides a large amount of data support and analysis means for medical research. Through the analysis of large-scale medical record data, some disease patterns, correlation and development trends that are difficult to find by traditional research methods can be found, which provides a new idea and direction for medical research and promotes the development of medical science. BRIEF DESCRIPTION OF DRAWINGS

[0058] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the application and serve to explain the principles of the application. In order to more clearly illustrate the technical schemes of the embodiments of the application, the drawings needed to be used in the embodiment description will be briefly introduced as follows. Obviously, those skilled in the art can obtain other drawings according to these drawings without any creative effort.

[0059] Figure 1 A flow chart of a method for quantifying the complexity and difficulty of a case according to an embodiment of the application.

[0060] The implementation, functional features and advantages of the application will be further described with reference to the embodiments and the accompanying drawings. Through the above drawings, the specific embodiments of the application have been shown, and more detailed descriptions will be given in the following. These drawings and textual descriptions are not intended to limit the scope of the concept of the application by any means, but to illustrate the concept of the application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0061] The exemplary embodiments will be described in detail herein with reference to the accompanying drawings. Unless otherwise indicated, the same numbers on different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with the application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the application as detailed in the appended claims.

[0062] It should be noted that, in the present document, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises... a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the element. Also, like-named components having a same or similar function are typically identified in the description for each embodiment of the application to ease the identification for the members of the public who can be aided by a better understanding of the application. This is for descriptive purposes only and is not intended to limit the scope of the application in any way.

[0063] It should be understood that the specific embodiments described herein are merely exemplary and do not limit the application.

[0064] First embodiment

[0065] Reference is made to Figure 1 The present application provides a method for quantifying case complexity and difficulty, which is based on disease diagnosis related grouping to calculate case complexity and difficulty using relative weights. The method comprises the following steps:

[0066] S10: Collect sample data and pre-process the sample data to extract valid medical record data.

[0067] Before analyzing the medical record data, data cleaning and preprocessing are required. This includes handling missing values, outliers, duplicate records, and inconsistent data formats. For example, there may be incorrect data entry in a hospital's electronic medical record system, such as recording "male" as "female", or using an incorrect format when recording the date of birth. Data cleaning and preprocessing are the basic steps to ensure the accuracy of analysis.

[0068] S20: Scale processing of the medical record data to identify disease patterns and trends in the medical record data.

[0069] For example, the amount of medical record data in a hospital is usually very large, which may contain millions or even billions of records. Processing such a large data set requires efficient algorithms and computing resources. For example, using a distributed computing framework (such as Apache Hadoop or Apache Spark) to accelerate data processing and analysis.

[0070] To understand the regularity of disease occurrence and treatment effect, complex statistical analysis and modeling are needed. For example, regression analysis is used to study the impact of certain factors (such as smoking, drinking or high blood pressure) on disease incidence, or survival analysis is used to evaluate the impact of different treatment options on patient prognosis.

[0071] S30: Based on the disease mode and trend, input the medical record data into the DRG grouper for grouping calculation, and each case enters a DRG group.

[0072] S40: Obtain the weight parameter of the case and standardize the parameter to calculate the complexity and difficulty of the case.

[0073] The embodiment can predict the severity of the disease or the treatment result according to the existing characteristics of the patient by constructing a medical record prediction model, identifying disease patterns and trends using machine learning technology, providing a reference for doctors to develop personalized treatment plans, assisting medical decision-making, and improving treatment effect and patient recovery rate. The technical solution of the present application provides a large amount of data support and analysis means for medical research, assists in studying disease patterns, correlation and development trends through analysis of large-scale medical record data, and improves user experience.

[0074] Optionally, the step of collecting sample data and preprocessing the sample data to extract valid medical record data includes at least one of the following:

[0075] Perform data cleaning process, remove duplicate data in the sample data, handle missing values in the sample data, and check and verify logical errors and inconsistencies in the data, and correct the error data.

[0076] Illustratively, there may be duplicate records in the medical record data. By checking the unique identifier of the record or comparing the key fields such as medical record number, patient name, date of birth, etc., identical records are identified and deleted to avoid duplicate analysis and ensure the uniqueness and accuracy of the data.

[0077] Illustratively, for missing values in the medical record data, appropriate processing methods should be selected according to the specific circumstances. If the missing values are few, the records containing missing values can be deleted; if the missing values are many, methods such as mean filling, median filling, and filling based on similar cases can be used to fill in. For example, for a missing check index of a certain disease, the mean of the index of other patients with the disease can be used to fill in.

[0078] Illustratively, check the logical errors and inconsistencies in the data, such as date format errors, gender mismatch with certain diagnoses or treatment measures, etc. Correct the error data according to business rules and common sense, and if the correct value cannot be determined, mark it as abnormal data and further investigate.

[0079] Data conversion is performed on data in different formats, numerical sample data of different dimensions and ranges are converted to a unified standard scale, continuous data and analog data are converted to discrete data according to the requirements of quantization processing, and classification data in medical records are converted to digital coding form.

[0080] Exemplarily, numerical data of different dimensions and ranges are converted to a unified standard scale. Common methods include Z-score standardization, Min-Max standardization, etc. For example, the vital signs data of patients (such as body temperature, blood pressure, heart rate, etc.) are standardized to facilitate subsequent analysis and model construction.

[0081] Exemplarily, for continuous data, it is converted to discrete data according to business requirements. For example, the age of a patient is divided into different age groups, such as 0-18 years old, 19-60 years old, and over 60 years old, which can facilitate classification statistics and analysis.

[0082] Exemplarily, the classification data in medical records, such as disease diagnosis name, surgery name, department name, etc., are converted to digital coding form. Dictionary coding, one-hot coding, etc. can be used to make the data better processed and analyzed by computers, and also facilitate data mining and application of machine learning algorithms.

[0083] Data integration processing is performed to load the medical record data from different sources into a unified database for integration, and to associate the data in different systems according to the patient's identity.

[0084] Exemplarily, the medical record data of a hospital may come from multiple different systems, such as electronic medical record system, testing system, imaging system, etc. The data from different sources needs to be integrated to obtain complete patient medical record information. Through the establishment of a data sharing platform or the use of ETL (Extract, Transform, Load) tools, the data scattered in various systems is extracted, cleaned and converted, and then loaded into a unified data warehouse.

[0085] Exemplarily, in the process of integrating multi-source data, the data in different systems needs to be associated according to the unique identification of the patient (such as medical record number, ID number, etc.), to ensure that the data of the same patient can be accurately corresponded and integrated together. For example, the electronic medical record data of a patient is associated with the test report data, imaging examination data, etc. to fully understand the patient's condition and treatment.

[0086] The original data source is called to verify the accuracy of the sample data, and the consistency of the medical record data in different records is checked.

[0087] For example, the accuracy of electronic medical record data is verified by comparing with the original paper medical records or other reliable data sources. Key data fields such as diagnosis results, treatment measures, etc. are checked in detail to ensure the accuracy and completeness of data entry.

[0088] For example, the consistency of medical record data between different modules and different records is checked. For example, the basic information of the patient should be consistent in the admission record, medical record and discharge summary; there should be logical consistency between diagnosis and treatment measures, such as the use of a certain antibiotic, which should have a corresponding infectious disease in the diagnosis.

[0089] For example, it is ensured that the medical record data meets the relevant laws and regulations, industry standards and internal regulations of the hospital. For example, the collection, storage and use of data should comply with the requirements of laws and regulations, and at the same time meet the medical record management norms and standards formulated by the health department.

[0090] Optionally, the process of processing the medical record data on a large scale to identify disease patterns and trends in the medical record data includes:

[0091] Learning feature variables related to a specific disease from the medical record data using natural language processing technology;

[0092] Machine learning models can automatically learn features related to a specific disease from medical record data, such as symptoms, laboratory test results, imaging findings, etc. By analyzing these features, the model can classify cases into different disease groups. For example, by analyzing the patient's complaints and doctor's notes using natural language processing technology, cases can be classified into respiratory diseases, cardiovascular diseases, etc.

[0093] Select and construct feature variables related to disease diagnosis and treatment effect, aggregate similar cases into multiple class clusters, and mine the association between feature variables in medical record data;

[0094] For example, in order to extract useful information from medical record data, feature engineering is needed. This includes selecting and constructing feature variables related to disease diagnosis and treatment effect. Association rule mining: using Apriori algorithm, etc. to mine the association between different disease diagnoses, symptoms, test results, treatment methods, etc. in medical record data. For example, it is found that patients with a certain underlying disease are more likely to develop another disease in a certain season, or there is an association between a certain treatment method and a certain complication, thereby summarizing potential disease patterns.

[0095] For example, using algorithms such as K-means clustering, similar cases can be aggregated into different clusters based on multiple characteristics in the medical record data, such as patient age, gender, disease diagnosis, treatment cost, etc. Each cluster may represent a specific disease pattern or patient population characteristics. By analyzing the characteristics and distribution of these clusters, the aggregation trend of the disease and the disease manifestation pattern of different types of patients can be identified.

[0096] Building a medical record prediction model, using machine learning models to identify disease patterns and trends in the medical record data.

[0097] For example, a model for predicting disease severity or treatment outcome can be built based on patient age, gender, symptoms, laboratory test results, and medical history, etc. Machine learning and artificial intelligence techniques can help identify disease patterns and trends in medical record data. For example, natural language processing (NLP) techniques are used to analyze free-text clinical notes to automatically extract key information such as disease diagnosis, type of surgery, and treatment effect.

[0098] For example, classification algorithms such as decision trees, support vector machines, neural networks, etc. take various features in the medical record data as input to classify cases into different disease types or disease stages. By training the model, the importance of different features to disease classification can be learned, and the relationship between disease patterns and related features can be discovered. For example, a neural network-based model is built to distinguish between different severity of cardiovascular diseases. The model can automatically learn the complex relationship between blood pressure, blood lipids, ECG indicators, and disease severity, thereby helping to identify the development trend and different patterns of the disease.

[0099] Optionally, the process of using natural language processing techniques to learn feature variables related to a specific disease from the medical record data includes:

[0100] Segmenting the text data in the medical record data into individual words and tagging each word with a part of speech;

[0101] Analyzing the grammatical structure of each sentence, determining the dependency relationship between words, building a syntax tree, and extracting key phrases related to the disease according to the syntactic structure;

[0102] For example, analyze the grammatical structure of a sentence, determine the dependency relationship between words, and build a syntax tree. For example, for the sentence "The patient developed a cough due to a cold", through syntactic analysis, it can be concluded that "the patient" is the subject, "due to a cold" is the predicate, and "cough" is the object, thus clarifying the structure of the sentence and the relationship between the components.

[0103] Exemplarily, according to the syntactic structure, a key phrase related to the disease is extracted. For example, the key phrase "high blood pressure history" is extracted from "the patient has a long history of high blood pressure".

[0104] Identify medical entities mentioned in the text data in the medical record data, and classify the identified medical entities;

[0105] Exemplarily, various medical-related entities mentioned in the text are identified, such as disease names, symptoms, drug names, and examination items. For example, in "the patient did an electrocardiogram examination, which showed myocardial ischemia, and was taking aspirin", it is identified that "electrocardiogram" is an examination item, "myocardial ischemia" is a disease name, and "aspirin" is a drug name.

[0106] Exemplarily, the identified entities are classified and labeled to distinguish different types of feature variables. For example, "cough" and "fever" are labeled as symptom entities, and "coronary heart disease" and "diabetes" are labeled as disease entities.

[0107] According to the medical characteristics, the classified and labeled medical entities are analyzed, and according to the preset statistical method, the importance of each word in the medical record data is calculated to extract feature variables;

[0108] Exemplarily, according to the knowledge and experience in the medical field, some rules are formulated to extract specific feature variables. For example, it is stipulated that a sentence containing the keywords "chest pain" and "chest tightness" can extract the feature variable "cardiovascular system symptoms".

[0109] Exemplarily, the role of each word in the sentence in terms of semantics is determined, such as agent, patient, time, and place. For example, in "the doctor performed surgery on the patient on October 10, 2023", "the doctor" is the agent, "the patient" is the patient, and "October 10, 2023" is the time.

[0110] Exemplarily, the extracted feature variables are fused with the medical knowledge graph, and the prior knowledge in the knowledge graph is used to further understand and enrich the semantic information of the feature variables. For example, the identified disease name is associated with the disease classification, cause, and symptom information in the medical knowledge graph, so as to more comprehensively understand the feature variables related to the disease. If the disease entity "pneumonia" is extracted, through the knowledge graph, it can be known that pneumonia has multiple types, and common symptoms include fever, cough, and sputum, which can be used as additional feature variables for supplementation.

[0111] The extracted feature variables are converted into vector representations by using word embedding technology, so as to analyze the importance of the feature variables.

[0112] Feature importance analysis techniques (e.g., feature weights, permutation importance, etc.) can be utilized to identify the most important features for disease classification. For example, analysis finds that symptoms such as "cough" and "dyspnea" are very important for classifying as respiratory diseases.

[0113] Illustratively, statistical methods such as Term Frequency-Inverse Document Frequency (TF-IDF) can be used to calculate the importance of each word or phrase in the entire medical record data, and select the ones with higher importance as feature variables. Useful features can be extracted from medical record data using techniques such as TF-IDF, Word Embedding, etc. These features can include symptoms, disease names, surgery types, drug names, etc. For example, if "dyspnea" appears frequently in medical records of respiratory diseases but rarely in medical records of other diseases, "dyspnea" can be an important feature variable related to respiratory diseases.

[0114] Optionally, the word embedding technique includes the following in the process of converting the extracted feature variables into vector representations:

[0115] A vocabulary is constructed based on the feature variables of the medical record data, and each vocabulary in the vocabulary is converted into a fixed-length vector using a pre-trained word embedding model;

[0116] Illustratively, a vocabulary is constructed from the pre-processed text data, listing all unique words. A word embedding model (e.g., Word2Vec, GloVe, fastText, etc.) is used to convert each word in the vocabulary into a fixed-length vector. These vectors can capture the semantic relationships between words. For example, the symptoms "cough" and "dyspnea" may be close to each other in the vector space because they often appear together in the text.

[0117] Illustratively, common word embedding models include Word2Vec, GloVe, FastText, etc.

[0118] For example, the Word2Vec model has two training methods: Skip-Gram and Continuous Bag-of-Words (CBOW). The Skip-Gram model predicts context words given a head word, while CBOW predicts the head word based on context words. For instance, for the sentence "The patient developed a cough," the Skip-Gram model predicts context words such as "patient," "developed," "developed," and "symptoms" based on the head word "cough"; the CBOW model predicts the head word "cough" based on these context words.

[0119] For example, the GloVe model is trained based on a global word co-occurrence matrix, which combines the advantages of matrix factorization and local context windows, enabling it to better utilize global statistical information in the corpus.

[0120] For example, the FastText model, based on Word2Vec, considers the internal structure of words and can provide better vector representations for some out-of-vocabulary (OOV) words. It may be more advantageous when dealing with some obscure professional terms in the medical field.

[0121] For example, taking the Skip-Gram model of Word2Vec as an example, the segmented medical record data is used as input, and the hyperparameters of the model are set, such as the dimension of the word vector, the size of the context window, and the number of iterations.

[0122] For example, the word vector dimension determines the size of the vector space represented by each word, typically ranging from tens to hundreds of dimensions, such as 100 dimensions. The context window size represents the range of context words considered when predicting the center word; for example, setting it to 5 means considering 5 words before and after the center word.

[0123] For example, during training, the model learns the vector representation of each word based on the input text data. By continuously adjusting the parameters of the vectors, the model can accurately predict context words or head words, thus making words with similar semantics close in the vector space. For instance, the vectors of "cough" and "phlegm," both related to respiratory symptoms, will be relatively close in space after training.

[0124] Exemplarily, each word or phrase in the extracted feature variable is mapped to a corresponding vector by the trained word embedding model. For example, the feature variable "history of hypertension", in which "hypertension" and "history" have corresponding vector representations in the word embedding model, can be combined (e.g., added, averaged, etc.) to obtain a vector representation of "history of hypertension".

[0125] Exemplarily, if the feature variable is a sentence or a longer text segment, the vectors of each word in it can be processed synthetically, such as using average pooling, maximum pooling, etc. to obtain a fixed-length vector to represent the feature variable. For example, for the sentence "patient presents with fever, headache, and cough", the vectors of the words "patient", "presents", "fever", "headache", "with", and "cough" can be obtained first, and then an average pooling method is used to obtain a vector representing the entire sentence.

[0126] According to the converted vectors of each word, the medical record is represented as a vector sequence to be input into the machine learning model for training or prediction.

[0127] Once the words are converted into vectors, the entire text (such as a medical record) can be represented as a vector sequence. This sequence can be input into a machine learning model for training or prediction.

[0128] Optionally, the step of identifying disease patterns and trends in the medical record data using a machine learning model and the previous steps include:

[0129] Based on the multiple class clusters and association relationships, a comprehensive application of Naive Bayes, Support Vector Machine, Random Forest, Neural Network is performed to conditionally judge the features in the medical record data, and the medical record data is divided into different disease categories.

[0130] A regression model is established to predict the relationship between the hospitalization days and the age of patients corresponding to different diseases.

[0131] Exemplarily, if the type of disease is to be predicted (such as distinguishing between bacterial and viral pneumonia), a classification model such as decision tree, support vector machine, naive Bayes, etc. can be selected. Taking the decision tree as an example, it gradually divides the data into different disease categories by conditionally judging the features in the medical record data.

[0132] Exemplarily, when predicting continuous variables such as disease severity score, hospitalization days, etc., regression models such as linear regression, ridge regression, etc. can be considered. For example, a regression model is established to predict the relationship between the hospitalization days of patients due to a certain disease and factors such as the age of the patient, underlying diseases, etc.

[0133] Optionally, the process of performing conditional judgment on the features in the medical record data based on the plurality of clusters and the association relationship, and comprehensively applying based on Naive Bayes, support vector machine, random forest, neural network, and dividing the medical record data into different disease categories comprises:

[0134] The model fusion technology is used to combine the prediction results of multiple models in a stacking manner to determine the final classification result.

[0135] For example, Naive Bayes, support vector machine, random forest, and neural network are used to train the medical record data. Each model can learn different patterns from the data. For example, Naive Bayes can learn the conditional probability between features, support vector machine can learn the hyperplane that separates different categories of data, random forest can learn the non-linear relationship between features, and neural network can learn more complex patterns.

[0136] For example, model fusion techniques such as voting, weighted averaging, and stacking can be used to combine the prediction results of multiple models to improve the accuracy of prediction. For example, for a new medical record data, each model can be used to make a prediction, and then a voting method can be used to determine the final prediction result.

[0137] Optionally, the process of using model fusion technology to combine the prediction results of multiple models in a stacking manner to determine the final classification result comprises:

[0138] Based on the features of the medical record data, Naive Bayes is used to preliminarily classify the medical record data to obtain a first classification result;

[0139] The first classification result is used as the input feature of the support vector machine, and other features of the medical record data are combined to perform a second classification on the medical record data to obtain a second classification result;

[0140] The first classification result, the second classification result, and other features in the medical record data are used as the input of part of the decision trees in the random forest to perform a third classification on the medical record data to obtain a third classification result.

[0141] For example, based on Bayes' theorem, assuming that the features are independent of each other, the probability of belonging to a certain disease category given the features is calculated, and the category with the highest probability is selected as the prediction result.

[0142] For example, a dataset is divided into a training set and a test set. For a medical record, the features include symptoms such as "cough", "fever", "dyspnea", etc. There are two classes: "respiratory disease" and "other diseases". The Naive Bayes model calculates the probability of the medical record belonging to "respiratory disease" and "other diseases" based on the occurrence probability of these symptoms in different disease classes in the training data. For example, when "cough", "fever", and "dyspnea" occur simultaneously in the training data, the probability of belonging to "respiratory disease" is 0.7, and the probability of belonging to "other diseases" is 0.3. Therefore, the medical record is classified into the "respiratory disease" category.

[0143] For example, during the training process of the second layer model, the feature set generated by the first layer model can be used as input to train the second layer model. The second layer model can be a logistic regression, decision tree, or any other suitable model.

[0144] For example, during the model fusion process, the prediction results of the first layer model can be input to the second layer model. The second layer model weights or votes these inputs to generate the final prediction result.

[0145] For example, assume the following medical record data:

[0146]

[0147] For example, the dataset is divided into a training set and a test set. Naive Bayes, support vector machine, random forest, and neural network models are trained using the training set. Each model will make predictions on the training set to generate a new feature set. For example, assume that the support vector machine model predicts that a certain case has a probability of 0.8 of being pneumonia and a probability of 0.2 of being a cold, then the generated feature set is [0.8, 0.2].

[0148] For example, the feature set generated by the first layer model is used as input to train the second layer logistic regression model. The prediction results of the first layer model are input to the second layer model. The second layer model weights or votes these inputs to generate the final prediction result. For example, if the second layer model judges that the probability of pneumonia is higher, then the final prediction result is pneumonia. In this way, the stacking method allows the combination of the prediction results of multiple models to improve the accuracy of prediction.

[0149] For example, using the support vector machine approach, in medical record data classification, a boundary is found that best distinguishes different disease categories. For example, taking "cardiovascular disease" and "non-cardiovascular disease" as an example, assuming that the characteristics of the medical record data include "blood pressure value", "blood lipid value", "electrocardiogram index", etc. The support vector machine model will find an optimal hyperplane in the multi-dimensional space formed by these characteristics. For example, when the "blood pressure value" is higher than a certain threshold, the "blood lipid value" is within a certain range, and the "electrocardiogram index" presents a certain specific form, these data points will be classified into the "cardiovascular disease" category, on one side of the hyperplane; while other data points that do not meet these conditions will be classified into the "non-cardiovascular disease" category, on the other side of the hyperplane.

[0150] For example, for the classification of "digestive system disease" and "non-digestive system disease", each decision tree in the random forest may make judgments based on different combinations of characteristics. For example, one decision tree may judge whether it is a "digestive system disease" according to "abdominal pain", "diarrhea" and other symptoms, and another decision tree may judge according to "gastroscopy results", "liver function indicators" and other characteristics. Finally, the results of all decision trees are combined to determine whether the medical record data belongs to "digestive system disease" or "non-digestive system disease" by majority voting. If most of the decision trees think that a medical record belongs to "digestive system disease", then the medical record will be classified into the "digestive system disease" category.

[0151] For example, taking "neurological disease" and "non-neurological disease" as an example, the input layer of the neural network can be features related to the nervous system, such as "headache", "dizziness", "numbness", "electroencephalogram results", etc. After processing and learning by the hidden layer, the output layer will give the probability that the medical record belongs to "neurological disease" and "non-neurological disease". For example, after training, the neural network outputs a probability of 0.8 for "neurological disease" and a probability of 0.2 for "non-neurological disease" when the characteristics of a medical record data are input. Then the medical record will be classified into the "neurological disease" category.

[0152] Exemplarily, first, the medical record data is preliminarily classified using Naive Bayes to obtain a preliminary result. Then, the result is taken as one of the input features of the support vector machine, and meanwhile, other original features are combined to allow the support vector machine to further classify the data. Then, the classification results of Naive Bayes and the support vector machine are taken as the input of part of the decision trees in the random forest, and are comprehensively judged together with other decision trees based on original features. Finally, the results of the previous three algorithms and the original features are taken as the input of the neural network, and through the learning and adjustment of the neural network, the final disease category division is obtained. In this way, the advantages of various algorithms can be fully utilized by comprehensively applying multiple algorithms, so as to improve the accuracy of disease category division.

[0153] Exemplarily, for a complex medical record data containing various symptoms, examination results and other features, Naive Bayes preliminarily judges it as "respiratory system disease" according to the probability of the occurrence of the symptoms. The support vector machine further confirms whether it meets the classification boundary of "respiratory system disease" by combining the symptoms and some examination indexes such as lung imaging results. Part of the decision trees in the random forest refers to the results of Naive Bayes and the support vector machine, and meanwhile, considers other factors such as the age and medical history of the patient to make a comprehensive vote. Finally, the neural network takes the results of the previous three algorithms and all original features as the input, and after complex learning and calculation, it finally determines that the medical record data belongs to the "pneumonia" category in the "respiratory system disease", because the neural network considers that the medical record data meets the characteristic pattern and probability distribution of "pneumonia" after comprehensively considering various factors.

[0154] Optionally, in the process of inputting the medical record data into the DRG grouper for grouping calculation based on the disease mode and trend, the process of each case entering a DRG group includes:

[0155] performing MDC large-class grouping according to the medical record front page data;

[0156] dividing each MDC into a disease type, and using MDCA to preliminarily group diseases;

[0157] subdividing each disease type into DRGs groups according to the age, gender and severity of complications;

[0158] verifying each DRGs group using big data, and adjusting the group according to the CV coefficient of variation;

[0159] grouping the cases according to the age, gender, neonatal admission weight, neonatal discharge weight, marriage, admission and discharge date, main diagnosis, secondary diagnosis and operation parameters in the medical record data.

[0160] Optionally, in the process of obtaining the weight parameters of the cases and standardizing the parameters to calculate the complexity and difficulty of the cases, the process includes:

[0161] wherein the weight parameters include: primary care proportion pA, mortality rate pB, average hospitalization day pC and expense pD, and the standardization calculation formula of each parameter is formula (1)-(4):

[0162] Primary care proportion pA = corresponding DRG group primary care proportion value / total sample primary care proportion (1);

[0163] Mortality rate pB = number of deaths in a certain DRG group / total number of people in the DRG group (2);

[0164] Average hospitalization day pC = average hospitalization day of the corresponding DRG group / average hospitalization day of the total sample (3);

[0165] Expense pD = average medical expense of the corresponding DRG group / average medical expense of the total sample (4);

[0166] According to the weight parameters, the weight value RW of the case complexity and difficulty is calculated, and the weight value RW calculation formula (5) is:

[0167] RW = pA*r1+pB*r2+pC*r3+pD*r4 (5)

[0168] wherein r1+r2+r3+r4 = 1, r1 represents the severity of the disease; r2 represents the risk degree of the disease; r3 represents the degree of human resource consumption; and r4 represents the degree of medical resource consumption.

[0169] The present application can automatically segment, tag word forms, and analyze syntax structures of the text in the medical record data through natural language processing technology, efficiently extract disease-related key information such as symptoms, disease names, and test results from a large amount of free-text clinical notes, avoid the tediousness and inaccuracy of manual extraction, and greatly improve the efficiency and accuracy of information extraction. It can accurately identify medical entities in medical record data and perform classification annotation, which helps to clearly define different types of feature variables and provides clear and accurate basic data for subsequent disease classification and analysis. For example, accurately distinguishing between disease entities and symptom entities can better grasp the manifestations and diagnostic basis of diseases.

[0170] The present application can discover potential association relationships between different feature variables in medical record data, such as the relationship between disease diagnosis, symptoms, test results, and treatment methods, through association rule mining. This helps doctors and researchers to better understand the occurrence and development mechanism of diseases, discover potential disease patterns, and provide references for disease prevention and treatment, such as discovering the association between a certain underlying disease and a specific complication.

[0171] The present application uses clustering analysis to aggregate similar cases into clusters, each cluster representing a specific disease pattern or patient population characteristic. By analyzing the characteristics and distribution of these clusters, the clustering trend of diseases and the disease manifestation patterns of different types of patients can be identified, helping doctors have a clearer understanding of the development trend of the disease and take preventive and response measures in advance.

[0172] The present application can comprehensively use machine learning algorithms such as Naive Bayes, Support Vector Machine, Random Forest, and Neural Network to classify diseases using various features in medical record data as input, which can fully utilize the complex information in the data and improve the accuracy and reliability of disease classification. The model can learn the importance of different features for disease classification, thereby more accurately dividing medical record data into different disease categories, providing support for accurate diagnosis of diseases.

[0173] The present application can construct a medical record prediction model to identify disease patterns and trends using machine learning techniques, which can predict the severity of the disease or the treatment outcome based on the existing characteristics of the patient, provide reference for doctors to develop personalized treatment plans, assist medical decision-making, and improve treatment effect and patient recovery rate.

[0174] The present application can integrate medical knowledge graph to associate the extracted feature variables with prior knowledge in the knowledge graph, which can further enrich and expand the semantic information of the feature variables, making the model's understanding of disease-related features more comprehensive and in-depth, and helping to discover some medical knowledge and rules hidden in the data.

[0175] The technical solution of the present application provides a large amount of data support and analysis means for medical research, through the analysis of large-scale medical record data, some disease patterns, correlation and development trend that are difficult to be found by traditional research methods can be found, which provides new ideas and directions for medical research and promotes the development of medical science.

[0176] The application develops a performance DRG grouper from the disease types in the clinical front line, and selects parameters from four aspects of criticality, complexity, occupied human resources and consumed medical resources to calculate the performance weight RW (Relative Weight) of the disease group, so as to solve the weight value quantization problem close to the clinic, reflecting the complexity and difficulty of the case. The application can predict the severity of the disease or the treatment result according to the existing characteristics of the patient by constructing a medical record prediction model and using machine learning technology to identify disease patterns and trends, provide a reference for doctors to develop personalized treatment plans, assist medical decision-making, and improve treatment effect and patient recovery rate. The technical solution of the application provides a large amount of data support and analysis means for medical research, and through the analysis of large-scale medical record data, some disease patterns, correlation and development trends that are difficult to find by traditional research methods can be found, providing a new idea and direction for medical research and promoting the development of medical science.

[0177] It should be noted that in the present application, step codes such as S10, S20, etc. are used, the purpose of which is to more clearly and briefly describe the corresponding content, and does not constitute a substantial limitation on the order. Those skilled in the art may perform S20 before S10, etc. when implementing, but these should be within the scope of protection of the present application.

[0178] In the embodiments of the system and storage medium provided by the present application, any of the above method embodiments can contain all the technical features, and the description and explanation content is basically the same as that of the above method embodiments, which will not be repeated here.

[0179] The embodiments of the present application also provide a computer program product, which comprises computer program code, and when the computer program code runs on a computer, the computer executes the method in various possible embodiments as above.

[0180] The embodiments of the present application also provide a chip, which comprises a memory and a processor, the memory is used to store a computer program, and the processor is used to call and run the computer program from the memory, so that the device installed with the chip executes the method in various possible embodiments as above.

[0181] It can be understood that the above scenarios are only examples and do not constitute a limitation on the application scenarios of the technical solutions provided by the embodiments of the present application. The technical solutions of the present application can also be applied to other scenarios. For example, those skilled in the art can know that with the evolution of system architecture and the appearance of new business scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0182] The above application example serial numbers are only for description, not representing the advantages and disadvantages of the embodiments.

[0183] The steps in the method of the embodiments of the present application can be adjusted in sequence, combined and deleted according to actual needs.

[0184] The units in the device of the embodiments of the present application can be combined, divided and deleted according to actual needs.

[0185] In the present application, for the same or similar term concept, technical solution and / or application scene description, generally only the first time is described in detail, and for the sake of brevity, the repeated description is generally not repeated. When understanding the technical solutions and the like of the present application, for the same or similar term concept, technical solution and / or application scene description and the like which are not described in detail in the rear, the relevant description of the previous can be referred to.

[0186] In the present application, the description of each embodiment has its own emphasis, and the part not described or recorded in a certain embodiment can be referred to the relevant description of other embodiments.

[0187] The technical features of the technical solutions of the present application can be combined arbitrarily, in order to make the description simple, the above-mentioned embodiments of each technical feature all possible combinations are not described, however, as long as the combination of these technical features does not exist contradictory, should be considered as the scope of the present application.

[0188] The above is only the preferred embodiment of the present application, and does not limit the patent scope of the present application, any equivalent structure or equivalent process transformation using the content of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for quantifying the complexity and difficulty of cases, characterized in that, The method for quantifying case complexity and difficulty is based on disease diagnosis-related grouping and uses relative weights to calculate case complexity and difficulty. The method includes the following steps: S10: Collect sample data and preprocess the sample data to extract valid medical record data; S20: Perform large-scale processing on the medical record data to identify disease patterns and trends in the medical record data; S30: Based on the disease patterns and trends, input the medical record data into the DRG grouper for grouping calculation, and each case enters a DRG group; S40: Obtain the weight parameters of the cases and standardize the parameters to calculate the case complexity and difficulty. The process of scaling up the medical record data to identify disease patterns and trends in the medical record data includes: Natural language processing techniques are used to learn characteristic variables related to specific diseases from the medical record data; Select and construct feature variables related to disease diagnosis and treatment outcomes, aggregate similar cases into multiple clusters, and mine the correlation between feature variables in the medical record data; Construct a medical record prediction model and use a machine learning model to identify disease patterns and trends in the medical record data; The process of learning disease-related feature variables from the medical record data using natural language processing technology includes: The text data in the medical record data is segmented into individual words, and each word is labeled with its part of speech; Analyze the grammatical structure of each sentence, determine the dependency relationships between words, construct a syntax tree, and extract key phrases related to the disease based on the syntactic structure; Identify medical entities mentioned in the text data of the medical record data, and classify and label the identified medical entities; Based on the medical characteristics, the medical entities are classified and labeled, and the importance of each word in the medical record data is calculated according to the preset statistical method in order to extract feature variables; Word embedding technology is used to convert the extracted feature variables into vector representations for importance analysis of the feature variables; The process of inputting medical record data into a DRG grouper for grouping calculation based on the disease patterns and trends, with each case entering a DRG group, includes: Grouping by MDC category based on the data on the medical record front page; Each MDC is assigned a disease category, and diseases are initially grouped using MDCA; Each disease is further subdivided into DRGs groups based on age, gender, and the severity of complications; Each DRG group was validated using big data, and the groups were adjusted based on the coefficient of variation (CV). Cases were grouped based on parameters in the medical record data, including age, sex, neonatal admission weight, neonatal discharge weight, marital status, admission and discharge dates, primary diagnosis, secondary diagnosis, and surgery. The process of obtaining the weight parameters of the cases and standardizing the parameters to calculate the complexity and difficulty of the cases includes: The weighted parameters include: the proportion of Level 1 nursing care (pA), mortality rate (pB), average length of stay (pC), and cost (pD). The standardized calculation formulas for each parameter are formulas (1)-(4): The proportion of Level 1 nursing care pA = the proportion of Level 1 nursing care in the corresponding DRG group / the proportion of Level 1 nursing care in the total sample (1); Mortality rate pB = number of deaths in a DRG group / total number of deaths in the DRG group (2); Mean length of stay pC = Mean length of stay of the corresponding DRG group / Mean length of stay of the total sample (3); Cost pD = Average medical cost per visit in the corresponding DRG group / Average medical cost per visit in the total sample (4); The weight values ​​RW for case complexity and difficulty are calculated based on the weight parameters, where the formula (5) for calculating the weight value RW is: RW = pA*r1+pB*r2+pC*r3+pD*r4 (5) Where r1+r2+r3+r4 = 1, r1: represents the severity of the disease; r2: represents the risk level of the disease; r3: represents the level of human resource consumption; r4: represents the level of medical resource consumption.

2. The method for quantifying the complexity and difficulty of a case according to claim 1, characterized in that, The step of collecting sample data and preprocessing the sample data to extract valid medical record data includes at least one of the following: Perform a data cleaning process to remove duplicate data from the sample data, process missing values ​​in the sample data, check for logical errors and inconsistencies in the validation data, and correct erroneous data. Data conversion is performed on data in different data formats, converting numerical sample data of different dimensions and ranges into a unified standard scale, converting continuous data and simulated data into discrete data according to the needs of quantitative processing, and converting classification data in medical records into digital coding form; Data integration processing is performed, loading the medical record data from different sources into a unified database for integration and consolidation, and associating data from different systems based on the patient's identity identifier; The original data source was retrieved to verify the accuracy of the sample data and to check the consistency of the medical record data across different records.

3. The method for quantifying the complexity and difficulty of a case according to claim 1, characterized in that, The process of converting extracted feature variables into vector representations using word embedding technology includes: A vocabulary is constructed based on the feature variables of the medical record data, and a pre-trained word embedding model is used to convert each word in the vocabulary into a fixed-length vector. The medical record data is represented as a vector sequence based on the vector of each word transformation, so as to be input into the machine learning model for training or prediction.

4. The method for quantifying the complexity and difficulty of a case according to claim 1, characterized in that, The step of using a machine learning model to identify disease patterns and trends in the medical record data, and preceding thereof, includes: Based on the multiple clusters and relationships, and by comprehensively applying Naive Bayes, Support Vector Machine, Random Forest and Neural Network, the features in the medical record data are conditionally judged, and the medical record data is divided into different disease categories. Establish a regression model to predict the relationship between the length of hospital stay and age for patients with different diseases.

5. The method for quantifying the complexity and difficulty of a case according to claim 4, characterized in that, The process of classifying the medical record data into different disease categories by comprehensively applying Naive Bayes, Support Vector Machine, Random Forest, and Neural Network based on the multiple clusters and associations to make conditional judgments on the features in the medical record data includes: Model fusion technology is used to combine the predictions of multiple models in a stacking manner to determine the final classification result.

6. The method for quantifying the complexity and difficulty of a case according to claim 5, characterized in that, The process of using model fusion technology, which combines the prediction results of multiple models in a stacking manner to determine the final classification result, includes: Based on the characteristics of the medical record data, Naive Bayes is used to perform a preliminary classification of the medical record data to obtain a first classification result. The first classification result is used as the input feature of the support vector machine, and combined with other features of the medical record data, the medical record data is classified a second time to obtain the second classification result; The first classification result, the second classification result, and other features in the medical record data are used as inputs to a partial decision tree in a random forest to perform a third classification on the medical record data and obtain the third classification result.

Images