Localization-oriented Disease Diagnosis Related Group Grouping And Prediction Algorithm Research

In recent years,Diagnosis Related Groups(DRGs)have been unsatisfactory in the evaluation of medical service providers and medical payment methods in different regions.Machine learning method is one of the important methods to study classification problems.Compared with the traditional DRGs grouping method,it has great advantages.Therefore,this paper used machine learning-based classification algorithms to study DRGs localization grouping and classification prediction.First of all,DRGs are grouped in localized applications.The types and numbers of diseases in each country or region are different,and the level of medical insurance funds and hospitals are different.In view of the problem of unsatisfactory application effects of the same DRGs in different regions,localization is proposed.DRGs K-Means grouping algorithm.Use the One-Hot coding method to numerically process the complications and comorbidities of each case patient,and propose a score algorithm for complications and comorbidities based on multiple linear regression to determine the resource consumption of each complication and comorbidity The magnitude of the impact transforms discrete and disordered complications and comorbidities into numerical features.With age,complications and comorbidities as grouping characteristics,the first page of medical records is divided into DRGs with similar clinical process and equivalent resource consumption using DRGs K-Means algorithm.Secondly,because the current DRGs grouping is determined based on the information on the home page of the medical record after the patient is discharged from the hospital,it reduces the treatment efficiency of the medical service provider in the medical diagnosis and treatment process.At the same time,the patient cannot understand the level of resource consumption of their own condition in time.To solve the above problems,a DRGs grouping prediction algorithm based on XGBoost is proposed.The SMOTE algorithm is used to process unbalanced data,and the Bayesian optimization algorithm is used to determine the parameters of the prediction model.Age,complications and comorbidities are used as features to realize DRGs grouping prediction.Finally,using the preprocessed data set,in the same experimental environment,the two algorithms are designed to compare experiments,and the multiple evaluation indicators are compared with the currently widely used related algorithms for verification.