CN113593694A

CN113593694A - Method for predicting prognosis of severe patient

Info

Publication number: CN113593694A
Application number: CN202110601994.2A
Authority: CN
Inventors: 胡安民; 李惠萍; 单智铭; 王炳森; 钟雄雄
Original assignee: Shenzhen Peoples Hospital
Current assignee: Shenzhen Peoples Hospital
Priority date: 2021-05-31
Filing date: 2021-05-31
Publication date: 2021-11-02

Abstract

The invention provides a method for predicting the prognosis of a critically ill patient, and belongs to the field of prognosis evaluation of critically ill patients. The method comprises the following steps: incorporating patient and hospital related data in structured language; screening out characteristic variables based on a random forest, and constructing a prediction model through regularized logistic regression, K nearest neighbor, a support vector machine, the random forest, extreme gradient lifting and a deep neural network algorithm; and (3) performing ensemble learning by using the prediction probabilities of different methods through a limit gradient lifting algorithm to finally predict whether the critically ill patient dies or not, and further calculating the occurrence probability of the critically ill patient. The invention utilizes the patient data and the characteristic information of the hospital as much as possible, carries out individualized prediction on the prognosis of the critically ill patient, weakens the prediction bias caused by a certain model, and increases the prediction accuracy.

Description

Method for predicting prognosis of severe patient

Technical Field

A method for predicting the prognosis of a critically ill patient, in particular to a method for predicting the prognosis of a critically ill patient after integrated learning based on a plurality of machine learning methods.

Background

Critically ill patients often have life-threatening organ or system dysfunction, and early assessment and positive treatment is critical to saving the patient's life. Critical patients have complex and rapid changes; it is difficult to assess the risk of death of a patient by subjective experience alone. At present, the severity of the illness state of critical patients is reflected clinically through some scoring systems, and the death prediction system has certain value, including simplifying an acute physiology scoring model, an acute physiology and chronic health condition scoring model, a death probability model and the like.

These models are built based on logistic regression methods and suffer from several drawbacks, including that (r) the models require a linear relationship between the predictor variables and the corresponding results. While fractional polynomials can be used to fit the non-linear relationship, this requires that the modeler identify an efficient representation of the predictor variables so that the model has the best predictive performance; ② the existence of the sharing will lead to the estimation of the regression coefficients in the LR model being unstable, so that the model coefficients are no longer interpretable. And the logic model is sensitive to multi-element collinearity data, missing data and unbalanced data and is difficult to realize high-order interaction. And fourthly, external verification finds that the performance of the existing model is not ideal. Mortality can be overestimated, for example, when these models are used to develop data beyond that of development.

In view of the above-mentioned drawbacks, the inventors of the present invention have finally obtained the present invention through long-term studies and time.

Disclosure of Invention

The purpose of the invention is: the method is based on an integrated learning model combining K nearest neighbor, a support vector machine, random forests, extreme gradient lifting and a deep neural network algorithm, predicts prognosis of critically ill patients by extracting characteristic attributes of critically ill patients in big data, and further calculates the occurrence probability of critically ill patients.

The technical method of the invention is as follows: a prognostic probability calculation method based on severe patient big data comprises the following steps:

step S1: data that may affect the prognosis of critically ill patients is included, including: basic information of the patient, vital signs and laboratory findings entered into the intensive care unit, the number of beds in the hospital and intensive care unit.

Step S2: and (4) screening characteristic variables based on a random forest algorithm and recording a characteristic variable data set 1.

Step S3: 6 different prediction models are sequentially constructed on the data set 1 through regularized logistic regression, K nearest neighbor, a support vector machine, random forests, extreme gradient lifting and a deep neural network algorithm.

Step S4: and 6 prediction models based on different algorithms respectively calculate prediction probabilities for severe patients of historical data, and a limit gradient lifting model of ensemble learning is constructed again based on the prediction probabilities to obtain a final prediction model.

The invention provides a critical patient prognosis prediction method based on statistical method and machine learning method integrated learning.

Compared with the prior art, the method better utilizes original real data, reduces prediction errors caused by certain model bias, and more accurately predicts the prognosis of severe patients with complicated illness.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below.

FIG. 1 is a block diagram of the prognosis prediction of severe patients based on big data according to the present invention.

Detailed Description

The above and further features and advantages of the present invention are described in more detail below with reference to the accompanying drawings.

As shown in FIG. 1, the present invention is a prognostic probability calculation method based on big data of critically ill patients, which is designed for more precise stratification of the disease of critically ill patients, and the method extracts the collected patient information data and laboratory test results from the hospital system and automatically collects the vital sign data of patients from the monitoring equipment of the intensive care unit.

And constructing a prediction model based on the original big data of the severe patients. Firstly, screening out characteristic factors influencing the prognosis of a patient based on a random forest method through a training sample.

And then, sequentially constructing prediction models based on the screened characteristic variable data sets through different statistical methods and machine learning algorithms, and outputting prediction probability values of prognosis predictions of the patient models.

And then, performing secondary ensemble learning on the predicted probability value of the patient through a limit gradient lifting algorithm to obtain a final prediction model.

Because the etiology of the severe patients is complex, the corresponding data characteristics are different, and in order to predict the variable value more accurately, different statistical methods and computer algorithms are adopted to predict the variable value, so that the application is more practical.

According to the method for predicting the prognosis of the critically ill patients based on big data ensemble learning, the ensemble learning prediction model is constructed by combining a statistical method and a computer algorithm, so that potential errors of a single prediction model are avoided, and the prognosis of local critically ill patients can be reflected more individually based on diagnosis and treatment levels of different hospitals.

The above examples are only for illustrating the technical idea and features of the present invention, and the purpose thereof is to enable those skilled in the art to understand the content of the present invention and implement the present invention, and not to limit the protection scope of the present invention. All equivalent changes or modifications made according to the technical solutions and concepts described above in the spirit of the present invention should be covered within the protection scope of the present invention.

Claims

1. A method for predicting the prognosis of a critically ill patient, which mainly comprises the following steps:

(1) screening characteristic variables;

(2) multi-model prediction;

(3) and (4) integrated learning of the extreme gradient lifting model.

2. The method of claim 1, wherein the multi-model prediction comprises: number regularization logistic regression, K nearest neighbor, support vector machine, random forest, extreme gradient boosting and deep neural network algorithm.