A depression assessment method, system and device based on GDS-9

Through the GDS-9 brief table based on project response theory and Riskslim algorithm, the problems of GDS-30's time-consuming, labor-intensive and reliable efficiency are solved, and fast and accurate depression assessment is achieved, which is suitable for clinical disease early warning.

CN118352066BActive Publication Date: 2024-11-19BEIJING NORMAL UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410458147.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-04-17
Publication Date
2024-11-19
Estimated Expiration
2044-04-17

AI Technical Summary

Technical Problem

The existing GDS-30 depression evaluation tools are too long, time-consuming and labor-intensive, and have too low reliability and efficiency. It is difficult to add weights to the simple scale such as GDS-15 and GDS-5 when calculating the total score of subject characteristics. It is necessary to distinguish subjects, and in clinical practice, it is necessary to quickly diagnose depressive traits.

Method used

Based on project response theory, the GDS-9 brief table was developed, and theta score was calculated using the Riskslim machine learning algorithm, and 9 key items were selected. The subjects' depression traits were calculated through computers, and the disease warning model was constructed, and parameters were optimized to improve measurement accuracy.

Benefits of technology

It has achieved accurate measurement of depression positive in the elderly with fewer questions, and the measurement effect has improved significantly, with high correlation with GDS-30, fast calculation speed, and can quickly judge the level of depression. It is suitable for clinical disease warning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118352066B_ABST
    Figure CN118352066B_ABST
Patent Text Reader

Abstract

The present invention belongs to the field of intelligent medical care, and specifically relates to a depression assessment method, system and device based on GDS-9. The method includes: obtaining the subject's answers to each item of GDS-9, the GDS-9 consists of 9 items, and the 9 items measure the subject's perception of the meaning of life, helplessness, mood, and worries about the past, present and future; calculating the subject's theta score according to the answers to the various items; judging the subject's depression level according to the theta score, if the score exceeds a threshold, it is judged as depression positive. The GDS-9 proposed in this application achieves accurate measurement of whether the elderly's depression is positive with fewer questions, and the measurement effect is good, which is a significant improvement compared with the existing simplified GDS.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and more specifically, to a depression assessment method, system, device, and computer-readable storage medium based on GDS-9. Background Art

[0002] At present, GDS-30 is too long, time-consuming and labor-intensive to administer, and with the development of the times, some items have shown their shortcomings. Although some researchers have tried to develop a shorter version of GDS, such as GDS-15 and GDS-5, most of these scales are developed based on classical measurement theory. When calculating the total score of the subjects' traits, it is difficult to distinguish the subjects by simply adding equal weights, which is extremely unreasonable. Therefore, it has caused a series of problems such as low reliability and validity in the testing process. Summary of the invention

[0003] At present, researchers mainly use IRT item response theory to solve this problem. Based on this, we developed a simplified version of the GDS scale through item response theory. However, item response theory uses Z scores to measure the characteristics of subjects, which makes it difficult for experimenters or clinicians to calculate the depressive traits of subjects through paper-and-pencil tests in a short period of time. However, in actual clinical practice, it is of great significance to give a diagnosis as soon as possible. Based on this, in 2017, a scholar used machine learning to develop the Riskslim algorithm, which can calculate the total score faster like CTT and estimate the weight of different questions. Based on this, this study first developed a GDS simplified form that can be calculated by computer based on IRT, and then used the Riskslim machine learning algorithm to develop a GDS simplified form, and compared the performance of the two simplified forms.

[0004] The present invention provides a depression assessment method based on GDS-9, which utilizes data processing and feature extraction to capture information in time series data, and proposes improvements to the parameters of the disease prediction model and performs effective optimization, thereby constructing a disease early warning model suitable for clinical use.

[0005] A depression assessment method based on GDS-9, comprising:

[0006] Obtaining the subjects' responses to various items of GDS-9, which consists of 9 items, and the 9 items measure the subjects' sense of meaning in life, sense of helplessness, mood, and worries about the past, present, and future;

[0007] Calculate the subject's theta score based on the answers to each of the items;

[0008] The depression level of the subject is determined based on the theta score, and if the score exceeds a threshold, it is determined to be depression positive.

[0009] Furthermore, the GDS-9 includes the following 9 items:

[0010] Do you feel like your life is empty?

[0011] Are you troubled by some thoughts that you can't get rid of in your mind?

[0012] Do you often feel isolated and helpless?

[0013] Are you often restless and upset?

[0014] Do you often worry about the future?

[0015] Do you often feel heavy and depressed?

[0016] Do you feel that there is no point in living like this?

[0017] Do you always worry about the past?

[0018] Do you feel that your situation is hopeless?

[0019] Preferably, the method for calculating the theta score of the subject according to the answers to the items is as follows: the contribution of the item to the theta score is calculated according to the answers to the items. If the subject answers yes to an item, the contribution of the item to the theta score is expressed as Y i =(θ-b i )×a i If the subject answers no to an item, the contribution of the item to the theta score is expressed as Y i =-b i ×(1-a i ), where Y i represents the contribution of the ith item to the theta score of GDS-9, b i represents the difficulty of item i, a i represents the discrimination of item i, θ represents the potential depressive traits of the subject; the theta score is based on θ = ∑Y i The calculated solution for θ;

[0020] Preferably, the threshold is the optimal cutoff value of the ROC curve for distinguishing between depression-positive and depression-negative based on the theta score of GDS-9.

[0021] Furthermore, the 9 items measuring depression of the GDS-9 were screened from the GDS-30.

[0022] Furthermore, the screening from GDS-30 is based on item response theory;

[0023] Optionally, the method for screening based on item response theory includes the following steps:

[0024] Using a model based on item response theory to fit the items of the GDS-30 on the screening data set to obtain a fitting model;

[0025] The items with insignificant S-χ2 values ​​in the fitted model were retained;

[0026] Items that satisfied the local independence assumption on the fitted model were retained;

[0027] retaining the items whose slopes of the fitted model are within the set slope range;

[0028] The items whose information function level of the fitted model reaches the set information function level value are retained.

[0029] Furthermore, the method for screening based on item response theory also includes fitting the item characteristic curve of GDS-30 using multiple models based on item response theory, selecting a model with good fitting effect as a fitting model used for screening items, and fitting the items of GDS-30 using the fitting model;

[0030] Preferably, the method for screening based on item response theory further comprises: using a validation set to validate the performance of the GDS-9, the validation method comprising: testing the internal reliability of the GDS-9, the internal reliability meeting expectations;

[0031] Optionally, a model based on item response theory is used to fit the GDS-9 using the validation set data to test the model fitting effect, and the performance of the GDS-9 is judged according to the fitting effect;

[0032] Optionally, the unidimensionality of the GDS-9 is tested, and the GDS-9 complies with the unidimensionality assumption;

[0033] Optionally, the local independence of the GDS-9 is checked, and the GDS satisfies the local independence;

[0034] Preferably, the satisfying of the local independence assumption means LD<10;

[0035] Preferably, the set slope interval is 1.35-3.98;

[0036] Preferably, the setting information function level value is 0.90-3.52.

[0037] Furthermore, the GDS-9 may also be replaced by the following 9 items:

[0038] Have you given up many activities and interests?

[0039] Are you afraid that something bad will happen to you?

[0040] Are you often restless and upset?

[0041] Do you often worry about the future?

[0042] Do you feel comfortable living now?

[0043] Do you often feel heavy and depressed?

[0044] Do you feel that life is full of vitality?

[0045] Do you feel like most people are much better than you?

[0046] Do you wake up happy in the morning?

[0047] Further, the screening from GDS-30 is performed using the Risklim algorithm to screen from GDS-30;

[0048] Preferably, the integer weighted total score of the 9 items is calculated by the Risklim algorithm, and the probability of the subject being positive for depression is determined and output according to the integer weighted total score, and if the probability of being positive exceeds a set value, it is determined to be positive for depression;

[0049] Preferably, the integer weighted total score of the GDS-9 is generated by the Risklim algorithm, and the formula is:

[0050] score=2×item2+1×item8+3×item11+3×item13+2×item15+3×item16+3×item21+2×item23+2×item27

[0051] Among them, score represents the integer weighted total score

[0052] Preferably, the formula for calculating the subject's depression positive probability based on the integer weighted total score is as follows:

[0053]

[0054] Among them, score represents the integer weighted total score, and P is the probability of positive diagnosis.

[0055] A depression assessment system based on GDS-9, characterized in that the system comprises:

[0056] Acquisition unit: used to obtain the subjects' responses to various items of GDS-9, wherein the GDS-9 consists of 9 items measuring depression, and the 9 items measure the subjects' sense of meaning in life, sense of helplessness, mood, and worries about the past, present, and future;

[0057] Calculation unit: Calculate the theta score of the subject according to the answers to each item;

[0058] Decision making unit: used to judge the depression level of the subject according to the theta score. If the score exceeds a threshold, it is judged as depression positive.

[0059] Preferably, the system further comprises a question answering unit, and the questions comprise the following 9 items:

[0060] Do you feel that life is empty?

[0061] Are you troubled by some thoughts in your mind that you can't get rid of?

[0062] Do you often feel isolated and helpless?

[0063] Do you often feel restless and upset?

[0064] Do you often worry about the future?

[0065] Do you often feel heavy and depressed?

[0066] Do you feel that living like this is meaningless?

[0067] Do you always worry about the past?

[0068] Do you feel that your situation is hopeless?

[0069] A depression assessment device based on GDS-9, characterized in that it includes: a memory and a processor; the memory is used to store program instructions; the processor is used to call the program instructions, and when the program instructions are executed, any one of the depression assessment methods based on GDS-9 is implemented.

[0070] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, any one of the depression assessment methods based on GDS-9 is implemented.

[0071] Advantages of this application:

[0072] 1. The GDS-9 proposed in this application can accurately measure whether the depression of the elderly is positive with fewer questions, and the measurement effect is good, which is a significant improvement compared with the existing simplified GDS;

[0073] 2. This application performs item screening based on the item response theory on the basis of GDS-30. The original theta scores of GDS-9 and GDS-30 obtained when measuring the same subject are highly correlated, and the correlation is very significant.

[0074] 3. This application uses a Risklim-based method to screen GDS-9. This algorithm can calculate the total score faster than CTT and also estimates the weights of different questions. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0076] Figure 1 is a schematic flow chart of a depression assessment method based on GDS-9 provided in an embodiment of the present invention;

[0077] Figure 2 is a schematic flow chart of a depression assessment system based on GDS-9 provided in an embodiment of the present invention;

[0078] Figure 3 is a schematic block diagram of a depression assessment device based on GDS-9 provided in an embodiment of the present invention;

[0079] Figure 4 It is a test information function curve of GDS-9 of a project provided by an embodiment of the present invention;

[0080] Figure 5 is the AUC of the GDS short-form scale generated by the Riskslim model provided in the embodiment of the present invention in the training and validation sets;

[0081] Figure 6 It is the ROC curve of diagnosis of different versions of GDS summary tables provided by the present invention on the validation set. DETAILED DESCRIPTION

[0082] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0083] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The sequence numbers of the operations, such as S101, S102, etc., are only used to distinguish between different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0084] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0085] Figure 1 : is a schematic flow chart of a depression assessment method based on GDS-9 provided in an embodiment of the present invention. Specifically, the method comprises the following steps:

[0086] S101: Obtain the subject's score on GDS-9, which consists of 9 items measuring depression, and the 9 items measure the subject's sense of meaning in life, sense of helplessness, mood, and worries about the past, present, and future;

[0087] 1. Subjects:

[0088] All participants in this study were from the Beijing Aging Brain Health Initiative (BABRI), an ongoing community-based cohort study in China that focuses on the prevention of dementia and early detection of cognitive impairment. Participants were urban non-demented older adults living in local communities in Beijing, all of whom were native Chinese speakers. Inclusion criteria were (i) at least 55 years old; (ii) no dementia or probable dementia according to DSM-IV-TR (American Psychiatric Association 2000); (iii) no history of neurological or systemic diseases known to affect brain function, including major vascular disease, head trauma, tumors, alcohol abuse, and epilepsy; and (iv) basic sociodemographic information. This study selected all the test data of the BABRI cohort up to December 2018, a total of 9531 data. According to the inclusion criteria, we excluded subjects who did not meet the criteria, and we identified 4109 BABRI participants who completed the paper version of the GDS-30. Then we excluded 361 participants with missing values, and finally included 3748 participants in the analysis.

[0089] In psychological and educational measurement, two measurement theories are often used to analyze the quality of test questions and papers, including classical test theory (CTT) and item response theory (IRT). CTT usually uses the sum of the scores of the subjects on each item as a reflection of the level of the subjects' potential traits, which is easy to understand, simple and practical. However, due to its simplicity, CTT has some insurmountable shortcomings, namely, its subject scores are too dependent on the test items, and the item parameters are too dependent on the subject samples. Relatively speaking, the item response theory (IRT) overcomes some limitations of CTT itself by providing estimates of the potential traits of items and subjects through a large number of mathematical operations, large samples, and strong assumptions. The specific advantages are as follows:

[0090] (1) The subjects’ abilities and item parameters are invariant;

[0091] (2) The subject ability parameters and item difficulty parameters are on the same scale and are comparable;

[0092] (3) It is more reasonable to provide different measurement errors for subjects with different ability levels;

[0093] 2. Methods:

[0094] In processing the GDS-30 questionnaire data, we first randomly divided 3748 participants into two samples of 1874 participants each, sample1 and sample2. The purpose of this is mainly to fit the IRT model or perform machine learning-based Riskslim training on sample1, and to test the IRT model fit of the simplified GDS short version scale and the diagnostic accuracy of Riskslim on sample2. We used the "mirt" package in R4.2.0 for IRT analysis, and determined that the measurement construct of GDS-30 is unidimensional through the unidimensionality test. By fitting the 1PLM, 2PLM, and 3PLM models of the unidimensional item response theory, the relative fit index (AIC, BIC) was used to determine the best fit model. The smaller the AIC and BIC values, the better the model fit. Then, the absolute fit index (M2) at the test level and the absolute fit index (S-χ2) at the item level were tested to determine whether the fit was good at the test and item levels (P value was not significant, p>0.05). If the fit was not good, the items that did not meet the local independence (LD>10) or the S-χ2 was large (P<0.05) were deleted until the absolute fit index at the test and item levels was not significant. Lord believed that when the slope parameter a of an item was greater than 1.35, it meant that the item had a high degree of discrimination (Lord, 1952). Therefore, we retained the items with a slope range greater than 1.35 as the final GDS short version. Next, we tested the absolute fit index and reliability of the GDS short version scale at the item and test levels on sample2. In addition, we also used the total score of the original GDS-30 (depressive traits measured by the original scale) as the criterion and conducted a correlation analysis with the theta score of GDS-9 (i.e., depressive traits measured by GDS-9). It should be noted that we use the EAP (Expecteda posteriori) method to estimate the subject's latent trait level (theta or θ). The scale based on IRT is different from the scale based on CTT: the questionnaire based on CTT is usually judged based on the total score of the subject's answers. The questionnaire or scale based on IRT theory measures the subject's latent trait θ, so the theta score is usually used to represent the subject's latent trait level measured on the questionnaire. Specifically, the estimation of the subject's traits is achieved through the fscores function in the mirt package.

[0095] 2.1 Introduction to IRT Model

[0096] IRT can analyze various types of test data, such as unidimensional and multidimensional test data, two-level scoring and multi-level scoring test data. For example, the Rasch model, one-parameter logistic model (1PL), two-parameter logistic model (2PL), three-parameter logistic model (1PL) and four-parameter logistic model (1PL) for two-level scoring data, and the graded response model (GRM), partial credit model (PCM) and generalized partial credit model (GPCM) for multi-level scoring data. It should be noted that in IRT, the Rasch model is the 1PL model, and the one-parameter and two-parameter here refer to the number of item parameters in the IRT model. The specific models are introduced as follows:

[0097] 2.1.1 IRT Model for Secondary Scoring Data

[0098] Let X ij represents the response of subject j (j = 1, 2, 3, ..., J) on item i (i = 1, 2, 3, ...,), then X ij There are two cases: 0 (wrong) and 1 (correct). In addition, let the item difficulty parameter be β i , the subject’s latent trait parameter is θ j In IRT analysis, the difficulty parameter and the subject's latent trait parameter generally follow a standard normal distribution, and their values ​​are comparable. That is, the relative size of the two parameters can be used to determine whether a question is difficult or easy for a subject of a certain level.

[0099] (1) Rasch model (1PL)

[0100] The Rasch model for dichotomous responses is sometimes called the simple Rasch model. The following formula represents the simple Rasch model. The item response function (IRF) was defined by Georg Rasch (1960).

[0101]

[0102] Obviously, the Rasch model (1PLM) assumes that subject j correctly answers item i, that is, X ij =1 with probability P ij The relative difficulty θ of item i for subject j is j -β i Decide.

[0103] (2) Two-parameter logistic model (2PLM)

[0104] The 2PLM proposed by Birnbaum (1957) is based on the difficulty of the project β i and discrimination α i The IRF of 2PLM is shown in formula (2):

[0105]

[0106] (3) Three-parameter logistic model (3PL)

[0107] Bimbaum (1968) introduced the project guessing parameter c based on 2PLM. i , 3PLM is proposed. Project guess parameter c i represents the probability that all subjects on item i guess the correct answer. 3PLM assumes that subject j correctly answers item i, that is, X ij =1 with probability P ij The relative difficulty θ of item i for subject j j -β i , item discrimination level α i and the project guessing degree c i Decided jointly that its IRF is as follows:

[0108]

[0109] (4) Four-parameter logistic model (4PL)

[0110] The 3PLM considers the guessing behavior of subjects with very low ability on the items, while the 4PL model takes into account the possible mistakes of subjects with high ability on the basis of the 3PL model, that is, it adds an item error parameter 1-u i It represents the possibility that a high-ability subject will make a mistake on item i. Its IRF formula is as follows:

[0111]

[0112] 2.1.2 IRT Model for Multi-level Scoring Data

[0113] Likewise, X ij represents the response of subject j (j = 1, 2, 3, .., J) on item i (i = 1, 2, 3, ..., I). In the multi-level scoring data, X ij There are i Let X ij =k, then k∈(0,1,2,…,m i ). In addition, the item threshold (difficulty) difficulty parameter of the multi-level scoring IRT model is β ih ,h∈(1,2,…,m i ), the subject’s latent trait parameter is θ j .

[0114] (1) Step-by-step scoring model (PCM)

[0115] The PCM model is a multi-level form of the Rasch model. It can handle response data with more than two categories. Its item parameter is only the difficulty parameter β. ih , IRF is as follows:

[0116]

[0117] P(X ikj =k|θ j ) represents the probability that subject j’s response category on item i is k. ih represents the threshold parameter for category k on item i, which can also be interpreted as the difficulty of scoring k on item i.

[0118] (2) Generalized Progressive Grading Model (GPCM)

[0119] The GPCM model is an extension of the PCM model (Muraki, 1992). The model allows the discrimination parameter c of the item to be freely estimated, and the IRF is as follows:

[0120]

[0121] (3) Graded Response Model (GRM)

[0122] The aforementioned PCM and GPCM models both model the probability of response of category k on item i, while the GRM model uses cumulative responses to model the probability of response of response category k or higher, and its cumulative response probability function is as follows:

[0123]

[0124] P(X ij ≥k|θ j ) represents the probability that the response category of subject j on item i is k or greater than k, β ikrepresents the threshold (difficulty) parameter for response category k on item i. The probability of a specific response in the GRM model is obtained by subtracting the adjacent cumulative probabilities:

[0125] P(X ij =k|θ j )=P(X ij ≥k|θ j )-P(X ij ≥k+1|θ j )

[0126] It should be noted that compared with the GPCM and PCM models, the threshold parameter of the GRM model assumes β ik ≥β i(k-1) , that is, the difficulty of getting a high score is always greater than the difficulty of getting a low score.

[0127] 2.2 RiskSLIM Screening

[0128] RiskSLIM (Risk-Calibrated Supersparse Linear Integer Model; Ustun & Rudin, 2017) is a machine learning algorithm that can efficiently find the most suitable logistic regression model with small integer weights and subject to custom constraints. Similar to previous work (Ustun et al., 2017; Ustun and Rudin, 2017), each model needs to comply with the constraints so that a fixed number of items are generated and a positive integer-valued score is generated, which is monotonic across different response levels. Since the dependent variable defined by Riskslim is a binary variable, we use the diagnostic criteria of GDS-30 (total score >10 is 1, and less than or equal to 10 is 0) as the dividing line. Referring to the source code of Ustun et al. (Python code for RiskSLIM is publicly available at https: / / github.com / ustunb / risk-slim), we use Python 3.11.1 to complete the item screening and model training of Riskslim on the training set (sample1). The model trained by sample1 is further tested using the AUC value on sample2. After obtaining the GDS-simplified version generated by IRT and Riskslim, we used the total score of the original GDS-30 scale as the criterion, diagnosed those with a total score greater than 10 as depression, and those with a total score less than 10 as non-depression. Then, we compared the diagnostic accuracy of the GDS-9 (IRT) developed in this study and the GDS-14 (Riskslim) with the highest AUC value with the GDS-4, GDS-5, and GDS-15 that were commonly used in previous studies in the validation set.

[0129] 3.Results

[0130] 3.1 IRT

[0131] 3.1.1 Assessing Unidimensionality

[0132] After reverse coding the 10 reverse-scored questions, we used the detect statistic in item response theory (detect = (sum(a*I))^2 / sum(I^2), where a represents the slope parameter and I represents the information function) to evaluate the unidimensionality of the latent traits measured by the GDS-30. The detect statistic is calculated based on the slope and information function of each item in the data. When the detect statistic value is less than 0.2, it indicates that the scale has a significant unidimensional structure. Our results show that the GDS-30 has a significant unidimensional structure (detect = -0.39 < 0.2).

[0133] 3.1.2 IRT Model

[0134] The three most commonly used two-level scoring IRT models in current research are the single-parameter logistic model (Rasch model; 1PLM), the two-parameter logistic model (2PLM), and the three-parameter logistic model (3PLM). Among them, 1PLM belongs to the Rasch model, which only contains the difficulty parameter, 2PLM contains two parameters of difficulty and discrimination, and 3PLM contains three parameters of difficulty, discrimination and guessing. We used 1PLM, 2PLM, and 3PLM to fit the item characteristic curve of GDS-30 in turn. It should be noted that when fitting 3PLM, in order to make the pseudo-guessing parameter smaller overall to better reflect the probability of the subject guessing the question correctly without knowing the answer, we limited the pseudo-guessing parameter through the prior distribution, that is, it was limited to a normal distribution with a mean of -1.1 and a standard deviation of 2. Then, by comparing the relative fitting indicators of the three models at the test level, we found that 2PLM had the best fitting effect, as shown in Table 1. Therefore, in the subsequent processing, we used 2PLM to fit all items. The formula of this model is as follows:

[0135]

[0136] Traverse 30 questions, j traverses n subjects, here, n = 1874.

[0137] Where Pij(θj) represents the probability of the jth subject with ability level (latent trait level; depression) θj to answer the i-th question correctly. αi and βi represent the discrimination parameter and difficulty parameter of the i-th question respectively, and θj is the j-th subject's latent trait level (depression).

[0138] Table 1. Relative fit indices of different IRT models

[0139]

[0140] 3.1.3 IRT Calibration and Item Selection

[0141] We used 2PLM to fit the item characteristic curves of the 30 items of GDS-30. By observing the absolute fit index of 2PLM at the test level, we found that the original GDS-30 scale had a poor fit (M2=3012.675, df=405, p<0.001, RMSEA=0.058). Previous studies generally believe that if the RMSEA is greater than 0.05, it means that the model fit is poor. Therefore, we further tested the local independence and S-χ2 of the GDS-30 items. This step was mainly achieved through the mirt, M2 and itemfit functions in the mirt package.

[0142] S-χ2 refers to the difference between the actual point and the estimated value during the item fitting process. The larger the value, the greater the error. Local independence is another important assumption in project response theory and is a prerequisite for project response theory analysis.

[0143] Therefore, in the subsequent analysis, we first eliminated the items with too large S-χ2 values ​​(significant P values) from large to small, and then eliminated the items that did not meet the local independence hypothesis (LD < 10), and finally ensured that the S-χ2 values ​​of all remaining items were not significant and met the local independence hypothesis (LD < 10). In this process, we eliminated a total of 12 items that did not meet the above requirements, and finally retained 18 items (Item 2, Item 3, Item 5, Item 6, Item 7, Item 8, Item 10, Item 11, Item 12, Item 13, Item 14, Item 16, Item 17, Item 18, Item 20, Item 22, Item 26, Item 29).

[0144] On this basis, we further estimated the item parameters of the test consisting of these 18 items, and estimated the corresponding discrimination a (slope parameter) and difficulty b (position parameter) for each binary response item. The results are shown in Table 2. The slope parameter measures the strength of the relationship between the item and the trait being measured - the stronger the relationship, the better the item. In addition, it is also used as an indicator of item quality - the higher the slope, the better it can distinguish items. The values ​​of slope parameter a in Table 2 range from 0.74 to 3.98. The position parameter b indicates the position where the probability of the item response being "yes" exceeds 50%. The difficulty b of the vast majority of items is moderately positive, which means that these are easier to distinguish subjects with depressive traits above the average level. The difficulty b of Item 14 is a high negative value, which means that almost all subjects will choose "yes" on this question.

[0145] Table 2. Estimates of slope parameters (a) and location parameters (b) for the 18 items in the GDS

[0146]

[0147] Consistent with a previous study (Dodeen & Al-Darmaki, 2016), we further screened the items based on the slope parameter a (discrimination) and the item information function. Lord believes that when the slope parameter a of an item is greater than 1.35, it means that the item has a higher degree of discrimination. Therefore, we retained items with a slope range of 1.35 to 3.98 (Item3, Item6, Item10, Item11, Item13, Item16, Item17, Item18, Item22), which have higher discrimination values ​​within the GDS range. The item information function (IIF) indicates the range in which an item or scale can best distinguish the subjects. The higher the information, the higher the accuracy and the greater the information that the test can provide (Reeve & Fayers, 2005). By drawing the information function curves of each item, we found that the information functions of the remaining 9 items were maintained at a high level (0.90-3.52) ( Figure 4 The total test information function curve can be shown in Figure 2. Therefore, we retained these 9 items as the final GDS brief form. The fitting effect of GDS-9 is better (M2=96.833, df=27, p<0.001, RMSEA=0.037). The total test information function curve that these 9 items can provide is shown in Figure 2. Figure 4 As shown in the figure, the simplified GDS-9 can provide a large amount of information when the depression level of the subjects is between 0-2, and is suitable for testing elderly people with moderate to high depression levels.

[0148] S102: Calculating the theta score of the subject according to the answers to each item;

[0149] S103: Determine the depression level of the subject according to the theta score, and if the score exceeds a threshold, determine that the subject is depression positive.

[0150] 3.1.4 Performance Verification of GDS-9

[0151] We used Sample 2 to further test the performance of the simplified GDS-9 based on Sample 1. We first used Cronbach's alpha coefficient to assess the internal reliability of the GDS-9 and found that the internal consistency of the GDS-9 was high (r = 0.838).

[0152] Secondly, we further verified the model fit of GDS-9 and 2PLM as well as the unidimensionality and local independence of the test on sample 2. The results showed that GDS-9 had a good model fit (M2=110.317, df=27, p<0.001, RMSEA=0.040<0.05), and the unidimensionality assumption (detect=-1.18<0.2) and local independence assumption (LD<10) were still met on sample 2.

[0153] In addition, we compared the IRT scores(s) obtained in the subjects using the original GDS-30 with the corresponding scores obtained using the GDS-9. We found that the original theta scores of the subjects in Sample 2 correlated very highly (r=0.87) and significantly (p<.001) with the theta scores of the same subjects using the GDS-9. This high correlation of the theta scores validates the performance of the GDS-9.

[0154] 3.2 Riskslim

[0155] 3.2.1 Riskslim item selection and scoring

[0156] Unlike the scale simplification method of IRT, Riskslim is a machine learning algorithm that uses modern optimization techniques to more rigorously develop short integer-scored screening scales to find a best-fitting logistic regression model with a fixed number of screening questions and the best integer score for each response option to predict clinical outcomes (Ustun et al., 2017). Table 3 shows the items selected in each Riskslim model of GDS in the training set. Considering that the current popular simplified scale items are less than 15 items, we only built Riskslim models with 1-15 items. In the 1-15 item model, almost all Riskslim selected item 16, which means that item 16 (Do you often feel heavy and depressed?) can well measure the depression level of individuals in the Riskslim model. We also found that when the maximum number of items was limited to 15, Riskslim only selected 14 items, which means that 14 items were able to achieve the best convergence of the model. It should be noted that Riskslim generates a unique weighted scoring method, which generates different integer weighted scoring methods when limiting different maximum number of items. For example, in the 9-item GDS, through the Riskslim algorithm, we generate the integer weighted total score of GDS-9 and the corresponding positive diagnosis probability, the formula is as follows:

[0157] score=2×item2+1×item8+3×item11+3×item13+2×item15+3×item16+3×item21+2×item23+2×item27

[0158]

[0159] If the participant's response pattern to the 9 items is 1 1 0 0 0 1 0 1 1, then the participant's total score = 10, and the corresponding positive diagnosis probability P = 0.88.

[0160] Table 3 GDS Items Selected in Riskslim Model

[0161]

[0162]

[0163] 3.2.2 Overall model fitting of the Riskslim algorithm

[0164] We will verify the 1-15 items GDS model obtained through the training set on the training set and the validation set. Consistent with previous studies (Ustun et al., 2017), we use the area under the ROC curve (AUC) as an indicator to judge the quality of the simplified GDS model. We found that when the number of GDS items exceeds 5, the AUC values ​​of the Riskslim model in the training set and the validation set are greater than 0.95, indicating that the 5-15 items of the GDS short-form scale generated by Riskslim has a higher accuracy in diagnosing geriatric depression. Figure 5 shown.

[0165] 3.3 Comparison of diagnostic accuracy of different versions of the GDS short form

[0166] In order to further compare the diagnostic accuracy of the short version of the GDS developed based on IRT and Riskslim and based on CTT, we compared the diagnostic accuracy of the GDS-9 (IRT) developed in this study and the GDS-14 (Riskslim) with the highest AUC value with the GDS-4, GDS-5, and GDS-15 commonly used in previous studies in the validation set. The results showed that the GDS-9 developed based on IRT had the highest diagnostic accuracy, the GDS-14 developed based on Riskslim ranked second, and the three GDS short forms developed based on CTT and commonly used in previous studies ranked last. For details, see Figure 6 . ROC curve analysis allows the determination of the optimal cutoff value by finding a balance between sensitivity and specificity. The generally accepted concept in previous studies is that the optimal cutoff value is located in the upper left corner of the ROC curve, which represents a point with high sensitivity and high specificity. In our study, we calculated the theta score corresponding to the point in the upper left corner of the ROC curve of the GDS-9 based on IRT simplification to be 0.54, which is the optimal cutoff value. This value is used as the threshold, that is, when the patient's score exceeds this threshold, they will be diagnosed as positive for depression.

[0167] In a specific embodiment, an elderly person answered nine questions in the GDS-9 simplified based on IRT. The elderly person's answers were: no, yes, no, yes, yes, yes, yes, no, no. The theta score of the elderly person's GDS-9 was calculated based on his answers. The calculation process was: (1) according to the two-point scoring of the question, the answer was converted into a score of 0 and 1; (2) if the subject answered yes (the score was 1), the contribution value of the question to the latent trait was calculated using the following calculation formula: Y i =(θ–b i )×a i If the subject answers no (score is 0), the following formula is used to calculate the contribution of the question to the potential trait: Y i=-b i ×(1–a i ). Among them, Y i represents the contribution value of the i-th question to the subject's latent trait θ, θ represents the subject's latent trait score. In the GDS-9 scale, the θ score measures the subject's level of depressive traits, b i represents the difficulty parameter of the i-th question, a i represents the slope parameter of the i-th question, that is, the discrimination parameter of the question; (3) Add up the contribution values ​​of all questions to obtain the subject's latent trait score, that is:

[0168] θ=∑Y i =Y1+Y2+Y3+…+Y9

[0169] Among them, according to the respondents’ answers, there are:

[0170] Y1=-b1×(1-a1)

[0171] Y2=(θ-b2)×a2

[0172] Y3=-b3×(1-a3)

[0173] Y4=(θ-b4)×a4

[0174] Y5=(θ-b5)×a5

[0175] Y6=(θ-b6)×a6

[0176] Y7=(θ-b7)×a7

[0177] Y8=-b8×(1-a8)

[0178] Y9=-b9×(1-a9)

[0179] According to Table 2, the difficulty parameters and discrimination parameters of the nine questions of GDS-9 are shown in Table 4:

[0180] Table 4. Estimation of discrimination parameters (a) and location parameters (b) for the nine items in GDS-9

[0181]

[0182] The corresponding a i and b i Substituting into θ=Y1+Y2+Y3+…+Y9, the solution of θ is: θ=0.468. This solution represents the depressive traits of the subject, that is, the theta score of the subject on GDS-9. Because 0.468 is less than the threshold value of 0.54, the elderly person is judged to be negative for depression.

[0183] In one embodiment, the above calculation process is written into a calculation program, and the computer program outputs theta scores according to the subjects' answers to each item of GDS-9, and then determines the depression status of the subjects.

[0184] Figure 2 It is a schematic flow chart of a depression assessment system based on GDS-9 provided in an embodiment of the present invention.

[0185] Acquisition unit: used to obtain the subject's answers to each item of GDS-9, the GDS-9 consists of 9 items, and the 9 items measure the subject's perception of the meaning of life, helplessness, mood, and worries about the past, present and future;

[0186] Calculation unit: Calculate the theta score of the subject according to the answers to each item;

[0187] Decision making unit: used to judge the depression level of the subject according to the theta score. If the score exceeds a threshold, it is judged as depression positive.

[0188] Preferably, the system further comprises a question answering unit, and the questions comprise the following 9 items:

[0189] Do you feel that life is empty?

[0190] Are you troubled by some thoughts in your mind that you can't get rid of?

[0191] Do you often feel isolated and helpless?

[0192] Do you often feel restless and upset?

[0193] Do you often worry about the future?

[0194] Do you often feel heavy and depressed?

[0195] Do you feel that living like this is meaningless?

[0196] Do you always worry about the past?

[0197] Do you feel that your situation is hopeless?

[0198] Figure 3 It is a schematic block diagram of a depression assessment device based on GDS-9 provided in an embodiment of the present invention.

[0199] A depression assessment device based on GDS-9, comprising: a memory and a processor;

[0200] The memory is used to store program instructions;

[0201] The processor is used to call program instructions, and when the program instructions are executed, the steps of the above-mentioned depression assessment method based on GDS-9 are implemented.

[0202] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned depression assessment method based on GDS-9 is implemented. A person skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the above-mentioned method embodiment, and will not be repeated here.

[0203] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0204] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0205] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0206] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0207] A person skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. The above-mentioned storage medium can be a read-only memory, a disk or an optical disk, etc.

[0208] The above is a detailed introduction to a computer device provided by the present invention. For a person skilled in the art, according to the concept of the embodiments of the present invention, there may be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A depression assessment method based on GDS-9, characterized in that: The method comprises: Obtain the subjects' responses to the various items of GDS-9, which consists of 9 items, and the 9 items measure the subjects' sense of meaning in life, sense of helplessness, mood, and worries about the past, present, and future; the GDS-9 consists of the following 9 items: Do you feel like your life is empty? Are you troubled by some thoughts that you can't get rid of in your mind? Do you often feel isolated and helpless? Are you often restless and upset? Do you often worry about the future? Do you often feel heavy and depressed? Do you feel that there is no point in living like this? Do you always worry about the past? Do you feel that your situation is hopeless? The theta score of the subject is calculated according to the answers to the items; the method for calculating the theta score of the subject according to the answers to the items is: the contribution of the item to the theta score is calculated according to the answers to the items, and if the subject's answer to an item is yes, the contribution of the item to the theta score is expressed as , if the subject answers no to an item, the contribution of the item to the theta score is expressed as ,in, represents the contribution of the ith item to the theta score of GDS-9, represents the difficulty of item i, represents the discrimination of item i, Indicates the potential depressive traits of the subject; the theta score is based on Calculated The solution of ; among them, and It is obtained by fitting the two-parameter logistic model to the items; The depression level of the subject is determined based on the theta score. If the score exceeds a threshold, it is determined to be depression-positive. The threshold is the optimal cutoff value of the ROC curve based on the theta score of GDS-9 to distinguish between depression-positive and depression-negative.

2. The depression assessment method based on GDS-9 according to claim 1, characterized in that: The 9 items measuring depression of the GDS-9 were screened from the GDS-30.

3. The depression assessment method based on GDS-9 according to claim 2, characterized in that: The screening from GDS-30 is based on item response theory.

4. The depression assessment method based on GDS-9 according to claim 3, characterized in that: The method for screening based on item response theory comprises the following steps: Using a model based on item response theory to fit the items of the GDS-30 on the screening data set to obtain a fitting model; The items with insignificant S-χ2 values ​​in the fitted model were retained; Items that satisfied the local independence assumption on the fitted model were retained; retaining the items whose slopes of the fitted model are within the set slope range; The items whose information function level of the fitted model reaches the set information function level value are retained.

5. The depression assessment method based on GDS-9 according to claim 4, characterized in that: The method for screening based on item response theory also includes fitting the item characteristic curve of GDS-30 using multiple models based on item response theory, selecting a model with good fitting effect as a fitting model for screening items, and fitting the items of GDS-30 using the fitting model.

6. The depression assessment method based on GDS-9 according to claim 4, characterized in that: The method for screening based on item response theory also includes using a validation set to validate the performance of the GDS-9, and the validation method includes: testing the internal reliability of the GDS-9, and the internal reliability meets expectations.

7. The depression assessment method based on GDS-9 according to claim 4, characterized in that: The GDS-9 was fitted using the validation set data using a model based on item response theory, the model fitting effect was tested, and the performance of the GDS-9 was judged based on the fitting effect.

8. The depression assessment method based on GDS-9 according to claim 4, characterized in that: The unidimensionality of the GDS-9 was tested and the GDS-9 met the unidimensionality assumption.

9. The depression assessment method based on GDS-9 according to claim 4, characterized in that: The local independence of the GDS-9 is tested, and the GDS-9 satisfies the local independence.

10. The depression assessment method based on GDS-9 according to claim 4, characterized in that: The local independence assumption is satisfied when LD < 10.

11. The depression assessment method based on GDS-9 according to claim 4, characterized in that: The set slope interval is 1.35-3.

98.

12. The depression assessment method based on GDS-9 according to claim 4, characterized in that: The set information function level value is 0.90-3.

52.

13. A depression assessment system based on GDS-9, characterized in that: The system comprises: Acquisition unit: used to obtain the subject's answers to each item of GDS-9, the GDS-9 consists of 9 items, and the 9 items measure the subject's perception of the meaning of life, helplessness, mood, and worries about the past, present and future; Calculation unit: Calculate the theta score of the subject according to the answers to the items; the method for calculating the theta score of the subject according to the answers to the items is: calculate the contribution of the item to the theta score according to the answers to the items, if the subject's answer to an item is yes, the contribution of the item to the theta score is expressed as , if the subject answers no to an item, the contribution of the item to the theta score is expressed as ,in, represents the contribution of the ith item to the theta score of GDS-9, represents the difficulty of item i, represents the discrimination of item i, Indicates the potential depressive traits of the subject; the theta score is based on Calculated The solution of ; among them, and The fitting of all items was obtained by using a two-parameter logistic model; Decision making unit: used to judge the depression level of the subject according to the theta score, and if the score exceeds a threshold, it is judged as depression-positive; the threshold is the optimal cutoff value of the ROC curve for distinguishing depression-positive and depression-negative based on the theta score of GDS-9; Answering unit: 9 items of GDS-9: Do you feel that life is empty? Are you troubled by some thoughts in your mind that you can't get rid of? Do you often feel isolated and helpless? Do you often feel restless and upset? Do you often worry about the future? Do you often feel heavy and depressed? Do you feel that living like this is meaningless? Do you always worry about the past? Do you feel that your situation is hopeless? 14. A depression assessment device based on GDS-9, characterized in that: include: Memory and processor; The memory is used to store program instructions; the processor is used to call program instructions, and when the program instructions are executed, the depression assessment method based on GDS-9 according to any one of claims 1 to 12 is implemented.

15. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the depression assessment method based on GDS-9 according to any one of claims 1 to 12 is implemented.