CN113053529A - Method for identifying and processing affective disorder - Google Patents

Abstract

The invention provides a method for identifying and processing affective disorder, which comprises the following steps of data sample selection, feature screening, feature sorting and model optimization. The invention carries out systematic analysis and modeling on ADE scale data and provides an evaluation method of model quality. The simplified ADE scale has the classification capability on normal, depression and biphase affective disorder equivalent to that of the original scale by optimizing and recombining the data items required by the function of the scale, but the number of actually used detection items is greatly reduced, and the effective auxiliary diagnosis is carried out on the affective disorder of the patient.

Description

Method for identifying and processing affective disorder

Technical Field

The invention relates to the technical field of medical prediction, in particular to a method for identifying and processing affective disorder.

Background

The ade (explicit person evaluation) scale is a standardized diagnostic tool for interviews of fixed patterns, and is revised based on the consensus of foreign experts. There is a very strong correlation between the ADE scale, which is widely used in the country, and affective disorders, and it is the systematic diagnostic assessment questionnaire that was originally used in the country.

With the aid of the ADE (active disease evaluation) scale, medical staff can know the medical history of a patient in detail, including the symptom manifestation of each episode, the past medical history, the age of onset, the use of drugs and therapeutic effects, comorbidities, personality traits, personal history and family history. Through quantitative scoring, the classification of normal, depression and bipolar affective disorder has good accuracy, and the sensitivity and specificity are all above 0.9.

However, the whole assessment process is relatively time-consuming and may have limited clinical application; in addition, one problem with the use of the ADE scale is that the amount of information that needs to be understood is relatively large, and its accuracy is closely related to the reliability of the source of the patient-related information. The patient's current disease severity may affect his or her knowledge of certain symptoms, leading to a shift in information.

The simplified ADE scale is obtained based on the theory and the method of medical big data processing, key items of the original scale are reserved, redundant items are deleted, the clinical auxiliary diagnosis efficiency is greatly improved on the premise that the misdiagnosis rate is not greatly increased, and the subsequent treatment process is further improved.

Disclosure of Invention

In order to solve the above problems, the present invention provides a method for identifying and processing affective disorder, which comprises the following specific technical solutions:

a method for recognizing and processing affective disorder is characterized in that: which comprises the following steps:

step 1: selecting a data sample; extracting effective samples available for modeling from a sample data set in the form of an ADE (active classifier evaluation) scale;

step 2: characteristic screening; preliminarily screening out the features with small variance from the feature set for constructing the sample data, and screening out the features with large correlation; deleting features related to disease course and result;

and step 3: sorting the importance of the features; calculating the importance weight of each feature in the feature set by using an mRMR (Minimum Redundancy and Maximum Redundancy Relevance) algorithm, and sequencing the importance weights from large to small to construct a feature importance sequence F;

and 4, step 4: the construction and optimization of the ADE reduced scale classification model comprises the following steps:

(4.1) combining Sequence Forward Selection (SFS) algorithm: extracting a feature set S with descending importance and sequentially ascending number from the feature importance sequence F;

(4.2) based on the sample data set, dividing the data set corresponding to the feature set S into a training set xtrin and a test set Xtest by using a random division mode;

(4.3) fitting an Xtrain data set on a training set by using a random forest algorithm, selecting model parameters by applying ten-fold cross validation in the modeling process, and constructing a classification model for normal, depression and biphase affective disorder;

(4.4) predicting data in the Xtest by using a training model, wherein the result is a classification score, and evaluating the performance of the classification model by combining an AUC (area Under cut) value;

(4.5) repeating the processes of 4.1-4.4 until the feature set S contains all features, and then selecting the feature set with the best performance from a series of feature sets as an optimal feature subset optimalset;

2. a method for affective disorder recognition and processing, as claimed in claim 1, wherein: in the step (4.1), the method comprises the following steps:

A. the initial of the characteristic set S is a null set, and the characteristic importance sequence F is not deleted;

B. adding the feature x with the maximum weight into the feature set S from the feature importance sequence F, and deleting the feature x which is already taken out from the F;

C. taking out a corresponding data set according to the characteristic set S, constructing a model on the divided training set by using a random forest algorithm, and calculating AUC (sensitivity, specificity) values of the corresponding test set;

D. repeating the step B, C until the feature importance sequence F is empty;

E. according to the steps, obtaining a feature set optimalset corresponding to the optimal AUC value;

F. and training according to the optimal feature set optimalset obtained in the process and the corresponding data set to obtain an optimal model.

3. A method for recognizing and processing affective disorder is characterized in that: the data set to be processed in step 1 is a data set containing the following information: present condition, life-long mania, life-long depression, time and frequency of episodes, and history of psychosis.

The method for identifying and processing the affective disorder has the following advantages:

according to the invention, ADE scale data related to mental diseases are used for carrying out systematic analysis and modeling, and a model evaluation method is provided, so that the diagnosis of mental diseases can be effectively assisted based on medical data through the model, and the inquiry efficiency of doctors is greatly improved on the premise of not increasing the misdiagnosis rate, thereby further improving the subsequent treatment process.

Drawings

FIG. 1 is a schematic workflow diagram of a method for identifying and processing affective disorders according to the present invention;

FIG. 2 is a feature importance ranking diagram of the present invention.

FIG. 3 is a graph of the relationship between the set of forward selected features of the sequence and the performance of the prediction model.

FIG. 4 is a ROC curve for the ADE reduced Scale of the present invention to differentiate depression from non-depression

FIG. 5 is a ROC curve for the ADE reduced scale of the present invention to distinguish between biphasic and non-biphasic

FIG. 6 is a ROC curve for the ADE reduced scale of the present invention to distinguish between normal and abnormal

Detailed Description

The following describes a method for identifying and processing affective disorder in detail with reference to the drawings and embodiments of the present invention.

The invention mainly carries out modeling based on ADE scale data in the electronic medical record, predicts the concrete types of the patients belonging to three mental diseases of normal, depression and biphase by using the information contained in the data, and carries out probabilistic description on the prediction result. The invention comprises a data processing flow for carrying out optimization modeling on medical data, and important methods and results for carrying out mental disease three-classification prediction, analysis, probabilistic analysis and the like. The invention combines medical data and a data mining method, is an innovation of combining medical data and a big data analysis method, fills the blank of the domestic mental scale data research to a certain extent, and has innovation in the aspect of carrying out mental disease classification prediction analysis by using ADE scale data.

The invention uses medical data derived from electronic medical record information collected in a mental disease scale database, and the main information in the data comprises ADE scale data and the mental disease category of a patient. As shown in fig. 1, the ADE reduced-scale classification prediction method based on medical data includes the following specific steps:

1. sample selection

The original dataset was dataset1, and a patient with severe data loss was removed from the dataset, which was now dataset 2.

2. Feature screening

For dataset2, feature screening is carried out, the variance of the feature value corresponding to each feature is calculated, the features with the variance close to 0 are removed, and the features with large correlation are screened out; deleting features related to disease course and result; the dataset at this point is dataset 3.

3. Feature importance ranking

Calculating the importance weight of each feature in the feature set by using an mRMR (Minimum Redundancy and Maximum Redundancy Relevance) algorithm, and sequencing the importance weights from large to small to construct a feature importance sequence F;

4, construction and optimization of ADE reduced scale classification model

(4.1) combining Sequence Forward Selection (SFS) algorithm: extracting a feature set S with descending importance and sequentially ascending number from the feature importance sequence F;

(4.2) based on the sample data set, dividing the data set corresponding to the feature set S into a training set xtrin and a test set Xtest by using a random division mode;

(4.3) fitting an Xtrain data set on a training set by using a random forest algorithm, selecting model parameters by applying ten-fold cross validation in the modeling process, and constructing a classification model for normal, depression and biphase affective disorder;

(4.4) predicting data in the Xtest by using a training model, wherein the result is a classification score, and evaluating the performance of the classification model by combining an AUC (area Under cut) value;

(4.5) repeating the processes of 4.1-4.4 until the feature set S contains all features, and then selecting the feature set with the best performance from a series of feature sets as an optimal feature subset optimalset;

example 1:

to verify the effectiveness of the ADE scale reduction method based on mental disease data of the present invention, the time of selection was 3 months from 2012, 9 central participants including 228 normal persons and 615 patients.

1. Data processing:

the method comprises two links of sample selection and feature screening, and adopts a data set with the form: each row is represented as a piece of patient information under examination, each column represents an aspect thereof, such as ID, physical examination information, psychiatric type, etc., and the data set format is as in Table 1. The original dataset contained 141 features, 19 of which were removed from the dataset, and finally 122 were left by data sample selection and feature screening.

TABLE 1 model basic information Table

The demographic information distribution thus obtained is shown in table 2:

TABLE 2 model demographic information distribution

2. Feature importance ranking

Sorting	ADE question item	Weight of
			1	65_ Current _ Severe mood (DysTHY)	15.60358631
2	n11_ manic life-long _ self-large (MSE _ EXP)	9.654495403
			3	21_ present _ past 1 year rise (P1Y _ ELV)	9.607021503
4	n29_ manic lifelong _ effort (up _ Energy)	9.085445726
			5	n55 time and number of episodes first manic age of treatment (mAge _ FT)	8.729160513
6	n27_ mania lifelong _ irritability (Easily innoyed)	8.381259716
			7	n10_ mania lifelong _ euphoria (MSE _ euPH)	8.216861866
8	23_ Current State _ irritation level (SEV _ IRT)	8.184733543
			9	20_ present _ upswelling degree (SEV _ ELV)	8.15059983
10	16_ present _ past two weeks of interest decline (P2W _ LoI)	8.08649208
			11	n24_ mania life-long _ adventure pleasure (RiskPL)	8.069190869
12	n18_ manic life-long _ language (MSE _ TLK)	7.823374303
			13	n30_ manic lifetime _ Spending (up _ speaking)	7.808619536
14	25_ present _ past two weeks anxiety (P2W _ Anx)	7.638252155
			15	17_ present _ degree of decline (SEV _ LoI)	7.611373104
16	n31_ mania lifelong _ Libido (up _ Libido)	7.523833422
			17	19_ present _ past two weeks upsurge (P2W _ Elv)	7.49774437
18	n17_ manic lifelong _ sleep need (MSE _ NfSLP)	7.428481244
			19	13_ present _ two weeks past Low (P2W _ Dep)	7.423776568
20	22_ present _ past 2 weeks irritability (P2W _ Irt)	7.400938891
			21	n25_ manic lifelong _ extraordinary achievement (exCOMPL)	7.358180672
22	n22_ manic lifelong _ psychomotor enthusiasm (MSE _ PMA)	7.265918966
			23	n 21-manic lifelong-aim with increased directional activity (MSE _ GDA)	7.158601284
24	n16_ manic lifelong _ self-assessment (MSE _ SlfEst)	7.148598986
			25	n19_ mania lifelong _ thought running (MSE _ FOI)	6.902996356
26	n14_ manic lifetime _ hospitalization (HSPTsized)	6.838146431
			27	14_ present _ degree of droop (Sev _ DEP)	6.785574511
28	27_ present _ past 1 year anxiety (P1Y _ Anx)	6.754640683
			29	55_ present _ thinking speed (P1W _ RT)	6.70226714
30	n32_ manic lifelong _ deviation concept (PI)	6.666430553
			…	…	…

TABLE 3 feature importance ranking

3. Modeling and optimization

Selecting a characteristic subset in an increasing mode through a sequence forward algorithm, wherein the characteristic subset is empty initially, selecting a characteristic with the largest weight, namely 65_ current state _ severe mood (DysTHY), from table 3, adding the characteristic subset, randomly disordering the data set on the basis of the data set corresponding to the characteristic subset, dividing the data set into 10 subsets, taking one subset as a test set and the rest as a training set in sequence, selecting a random forest algorithm to train and test a model, and obtaining corresponding sensitivity, specificity index and AUC value;

adding a feature subset with the second weight, namely n 11-mania lifelong-self-large (MSE _ EXP), into the feature subset, randomly disordering the data set according to the same sequence on the basis of the data set corresponding to the feature subset, dividing the data set into 10 subsets, taking one subset as a test set and the rest as a training set in sequence each time, and selecting a random forest algorithm to train and test the model to obtain corresponding sensitivity, specificity index and AUC value;

repeating the above processes until the feature subset contains all the features, ending the cycle to obtain a series of model parameters and performance indexes, and obtaining the optimized result from each extreme point.

4. Analysis of results

The optimal random forest algorithm parameters are as follows:

random forest algorithm	Classification Performance
		Number of decision trees	500
The number of features each decision tree contains	50
		Error rate	11.42％

TABLE 4 random forest model Performance indices

As can be seen from FIG. 3, when the feature subset contains 8 quantity table entries, the simplified model performance reaches the first extreme point, and the sensitivity and specificity indexes are close to the results obtained by all the entries of the ADE table; the simplified model then reaches an optimum extremum when the feature subsets contain 34, 67 terms, respectively. The classification performance index sensitivity and specificity statistics of the different feature subsets BDCC8, BDCC34 and BDCC67 are shown in the following tables 5, 6 and 7. The ROC curves are shown in fig. 4, 5, 6 for the three types of depression, biphase and normal and for the different subsets of features that reach extrema.

	ADE	BDCC 8	BDCC 34	BDCC 67
					Sensitivity of the probe	0.8985	0.8969	0.9147	0.9009
Degree of specificity	0.9496	0.9320	0.9312	0.9477

TABLE 5 Depression Classification Performance comparison of different feature subsets

	ADE	BDCC 8	BDCC 34	BDCC 67
					Sensitivity of the probe	0.9258	0.9433	0.9249	0.9319
Degree of specificity	0.8802	0.8556	0.8835	0.8841

TABLE 6 biphase Classification Performance comparison of different feature subsets

	ADE	BDCC 8	BDCC 34	BDCC 67
					Sensitivity of the probe	0.7395	0.6806	0.7009	0.7364
Degree of specificity	0.9566	0.9838	0.9667	0.9621

TABLE 7 comparison of Normal Classification Performance for different feature subsets

And (4) supplementary notes: regarding classification problems some index explanations, for a two-classification problem, two classes are defined as a positive class and a negative class, respectively, each object in the positive class is called a positive instance, and each object in the negative class is called a negative instance. In predicting disease, the disease is positive; normally negative. There are four cases when a classification model is used to predict a test sample, if an instance is a positive class and is predicted to be a True class (TP), and if an instance is a negative class, is predicted to be a positive class, it is called a False positive class (FP). Accordingly, if an instance is predicted as a negative class, called True Negative (TN), then a positive instance is predicted as a negative class, and False Negative (FN).

TP: positive examples are predicted to be positive type numbers;

FN: positive instance prediction is negative class number;

FP: the number of negative instances predicted as positive classes;

TN: the number of negative instances predicted as negative classes;

sensitivity (sensitivity): the correct prediction in positive class is the proportion of instances of positive class, i.e., TP/(TP + FN) specificity (specificity): example proportion of negative classes correctly predicted as negative classes, i.e. TN/(TN + FP)

Positive Predictive Value (PPV): in the case of the positive class, the proportion of the positive case is TP/(TP + FP).

And a one-to-many method is adopted to expand the two-classification into the multi-classification.

The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the claims of the present invention.

Claims (3)

1. A method for recognizing and processing affective disorder is characterized in that: which comprises the following steps:

step 1: selecting a data sample; extracting effective samples available for modeling from a sample data set in the form of an ADE (active classifier evaluation) scale;

step 2: characteristic screening; preliminarily screening out features with small variance from a feature set for constructing sample data, and deleting features related to disease course and results;

and step 3: sorting the importance of the features; calculating the importance weight of each feature in the feature set by using an mRMR (Minimum Redundancy and Maximum Redundancy Relevance) algorithm, and sequencing the importance weights from large to small to construct a feature importance sequence F;

and 4, step 4: the construction and optimization of the ADE reduced scale classification model comprises the following steps:

(4.1) combining Sequence Forward Selection (SFS) algorithm: extracting a feature set S with descending importance and sequentially ascending number from the feature importance sequence F;

(4.2) based on the sample data set, dividing the data set corresponding to the feature set S into a training set xtrin and a test set Xtest by using a random division mode;

(4.3) fitting an Xtrain data set on a training set by using a random forest algorithm, selecting model parameters by applying ten-fold cross validation in the modeling process, and constructing a classification model for normal, depression and biphase affective disorder;

(4.4) predicting data in the Xtest by using the trained classification model, wherein the result is a classification score, and evaluating the performance of the classification model by combining an AUC (area Under cut) value;

(4.5) repeating the processes of 4.1-4.4 until the feature set S contains all features, and then selecting the feature set with the best performance from a series of feature sets as an optimal feature subset optimalset.

2. A method for affective disorder recognition and processing, as claimed in claim 1, wherein: in the step (4.1), the method comprises the following steps:

A. the initial of the characteristic set S is a null set, and the characteristic importance sequence F is not deleted;

B. adding the feature x with the maximum weight into the feature set S from the feature importance sequence F, and deleting the feature x which is already taken out from the F;

C. taking out a corresponding data set according to the characteristic set S, constructing a model on the divided training set by using a random forest algorithm, and calculating AUC (sensitivity, specificity) values of the corresponding test set;

D. repeating the step B, C until the feature importance sequence F is empty;

E. according to the steps, obtaining a feature set optimalset corresponding to the optimal AUC value;

F. and training according to the optimal feature set optimalset obtained in the process and the corresponding data set to obtain an optimal model.

3. A method for affective disorder recognition and processing, as claimed in claim 1, wherein: the data set to be processed in step 1 is a data set containing the following information: present condition, life-long mania, life-long depression, time and frequency of episodes, and history of psychosis.

Images