Large-scale Screening for Cognitive Impairment Based on MemTrax

Abstract

The present invention proposes a large-scale screening technology for cognitive impairment based on MemTrax. In the first aspect, a model construction method is proposed, which simultaneously reflects two important factors, education and age, in the model; and, the concept of a comprehensive score is introduced to balance the bias that may be caused by simply looking at the correct rate of the test or the reaction time of the test; and, the GAMLSS modeling method is introduced to present the changing trend of cognitive function with age in a non-linear manner and supports parameter tuning to select the best model. The simultaneous introduction of education and age is the first time in the application involving MemTrax; the introduction of the comprehensive score is the first time in the norm based on MemTrax; the introduction of the GAMLSS modeling method is the first time in the cognitive field, and all are creatively used. In the second aspect, a large-scale screening device for cognitive impairment is proposed, and a standard for evaluating the trade-off phenomenon is proposed. When the conclusions given by the two indicators of the correct rate and the reaction time conflict, the preferred parameters are recommended or the comprehensive score is used for re-identification.

Description

Technical Field

[0001] The present invention relates to the technical field of medical assisted diagnosis, and particularly to the technical field of large-scale screening for cognitive impairment. Background Art

[0002] The present invention is mainly based on two aspects of background. First, dementia has become the seventh leading cause of death globally. The annual total treatment cost of Alzheimer's disease alone (the most important type of dementia, accounting for 60-70% of the cases) accounts for as high as 1.47% of China's GDP. The huge number of patients and the heavy social and family burdens have made it one of the biggest public health challenges currently and in the future. It is very important to intervene in a timely manner at the early stage of cognitive impairment. Early prevention, early identification, and early intervention are the fundamental ways to prevent and control dementia. Second, after more than thirty years of development, the value of the MemTrax Memory and Cognitive Assessment Software (hereinafter referred to as MemTrax) as a cognitive assessment tool has been generally recognized by the medical community, and it has been increasingly widely used due to its simple and easy-to-use features. The MemTrax Memory and Cognitive Assessment Software is a computerized cognitive assessment tool that relies on visual cues (observing, recognizing, and remembering consecutive images) and the use of hand functions (quickly clicking with the hand to recognize recurring images) as signal inputs, and the computer provides memory and cognitive assessments through algorithms.

[0003] It is very important to understand which types of cognitive changes are part of normal aging and which types of changes may indicate the occurrence of brain diseases. Therefore, using MemTrax to screen for cognitive impairment has attracted research interest. Currently, there is no patent on methods related to screening high-risk populations for cognitive impairment based on MemTrax, but relevant literature has been published, as follows:

[0004] Existing Solution 1: Professors from Stanford University in the United States, MemTrax Company, and HappyNeuron Company in France collaborated to collect data from more than 15,000 people and established a French norm (Ashford JW, Tarpin-Bernard F, Ashford CB, Ashford MT. A Computerized Continuous-Recognition Task for Measurement of Episodic Memory. Journal of Alzheimer's Disease 2019; 69(2): 385-399.). In this study, it was assumed that age-cognition showed a linear or quadratic relationship. With age as the independent variable x and the error rate or reaction time as the dependent variable, models were established separately by gender. The model established with the error rate as the dependent variable is shown in Table 1, where STD is the standard deviation; x is the age; y is the error rate; R 2is the coefficient of determination. The model established with the reaction time as the dependent variable is shown in Table 2, where STD is the standard deviation; x is the age; y is the reaction time; R 2 is the coefficient of determination.

[0005] Table 1: Model of error rate varying with age

[0006] Linear regression <![CDATA[R 2 > Quadratic regression <![CDATA[R 2 > Male Mean y = 0.0398x + 3.2782 0.5645 <![CDATA[y = 0.0004x 2 - 0.0044x + 4.3018]]> 0.5819 +1STD y = 0.0561x + 6.9124 0.4399 <![CDATA[y = 0.0007x 2 - 0.018x + 8.6283]]> 0.4591 +2STD y = 0.0723x + 10.547 0.3691 <![CDATA[y = 0.001x 2 - 0.0316x + 12.955]]> 0.3882 Female Mean y = 0.0444x + 3.3423 0.5048 <![CDATA[y = 0.001x 2 -0.0569x + 5.6902]]> 0.5706 +1STD y = 0.0646x + 6.8173 0.3794 <![CDATA[y = 0.0015x 2 - 0.0984x + 10.596]]> 0.4398 +2STD y = 0.0848x + 10.292 0.3199 <![CDATA[y = 0.0021x 2 - 0.1399x + 15.501]]> 0.3782

[0007] Table 2: Model of reaction time varying with age

[0008] Linear regression <![CDATA[R 2 > Quadratic regression <![CDATA[R 2 > Male Mean y = 0.0021x + 0.8033 0.7030 <![CDATA[y = 4E-05x 2 -0.0023x + 0.9035]]> 0.7804 +1STD y = 0.0028x + 0.9249 0.6288 <![CDATA[y = 5E-05x 2 -0.002x + 1.0367]]> 0.6758 +2STD y = 0.0035x + 1.0465 0.5489 <![CDATA[y = 5E-05x 2 -0.0018x + 1.1698]]> 0.5802 Female Mean y = 0.0026x + 0.7731 0.7844 <![CDATA[y = 5E-05x 2 -0.0029x + 0.8992]]> 0.8721 +1STD y = 0.0034x + 0.8824 0.7291 <![CDATA[y = 6E-05x 2 -0.0034x + 1.0407]]> 0.8016 +2STD y = 0.0043x + 0.9918 0.6719 <![CDATA[y = 8E-05x 2 -0.0039x + 1.1823]]> 0.7339

[0009] Existing Solution 2: The Bergeron MF team uses deep machine learning and the prediction of users' cognitive health based on MemTrax and users' health and medical history information (Bergeron MF, Landset S, Tarpin-Bernard F, Ashford CB, Khoshgoftaar TM, Ashford JW. Episodic-Memory Performance in Machine Learning Modeling for Predicting Cognitive Health Status Classification. Journal of Alzheimer's Disease 2019; 70(1): 277-286.). At the same time, the Bergeron MF team also uses MemTrax, demographic data, and medical history information from memory clinic patients and machine learning to predict mild cognitive impairment (MCI) (Bergeron MF, Landset S, Zhou X, Ding T, Khoshgoftaar TM, Zhao F and others. Utility of MemTrax and Machine Learning Modeling in Classification of Mild Cognitive Impairment. Journal of Alzheimer's Disease 2020; 77(4): 1545-1558.).

[0010] However, both of these existing solutions have serious drawbacks. In the existing solution 1, first of all, the French norm established by the Ashford team assumes a linear or quadratic relationship between age and cognition. However, the cognitive function shows a non-linear change trend with age but not necessarily a quadratic relationship. Therefore, the model established for the change of cognition with age does not match the actual situation well. Secondly, the French norm is modeled separately by gender stratification, but gender as a risk factor for Alzheimer's disease is still controversial. Thirdly, when the evaluation results of the correct rate and reaction time are inconsistent, no solution is given in this study. For the above reasons, the accuracy of the existing solution 1 is relatively low and it is not suitable for popularization among a large-scale population. In the existing solution 2, the Bergeron MF team uses machine learning to predict users' cognitive health and mild cognitive impairment (MCI) based on MemTrax, users' basic demographic data and medical history data. This study is quite different from the technical route of the present invention, and the established model is relatively complex. Ordinary users cannot know their own cognitive level after taking the home test. Therefore, it is also not suitable for large-scale population screening.

[0011] In summary, there is currently a lack of MemTrax-based cognitive impairment screening technology suitable for popularization among a large-scale population. Summary of the Invention

[0012] The following are the explanations of the abbreviations involved in the present invention:

[0013] MTx-RT: Reaction time in the MemTrax test results

[0014] MTx-%C: Correct rate in the MemTrax test results

[0015] MTx-Cp: Comprehensive score of the MemTrax test results (result balancing reaction time and correct rate)

[0016] The object of the present invention is to provide a MemTrax-based cognitive impairment screening technology suitable for popularization among a large-scale population. For this purpose, a method for constructing a model for large-scale screening of cognitive impairment based on MemTrax is proposed. At the same time, two important factors, education and age, are reflected in the model. And in the way that after classifying by education level, the corresponding relationship between age and MemTrax test data is obtained to reflect the relationship between these two important factors of education and age and cognitive impairment; in addition, the concept of comprehensive score is introduced, that is, a score is given by comprehensively reflecting the performance of reaction time and accuracy rate, balancing the bias that may be caused by simply looking at the accuracy rate or the reaction time of this test; in addition, the GAMLSS (Generalized Additive Models for Location, Shape and Scale) model is introduced to present the changing trend of cognitive function with age in a non-linear manner and support parameter tuning to select the best model. At the same time, the introduction of education and age is the first time in the application related to MemTrax; the introduction of comprehensive score is the first time in the norm based on MemTrax; the introduction of GAMLSS modeling method is the first time in the cognitive field, all of which are creatively used. In the cognitive impairment large-scale screening device applying the model constructed according to the foregoing model construction method, a standard for evaluating the trade-off phenomenon is proposed, and a method for recommending the preferred parameters or re-determining using the comprehensive score when the conclusions given by the two indicators of accuracy rate and reaction time conflict is given.

[0017] In a first aspect, an embodiment of the present invention proposes a method for constructing a model for large-scale screening of cognitive impairment based on MemTrax. The model construction method includes the following steps:

[0018] Data collection, data collation and data analysis.

[0019] The data collection is used to collect the data required for constructing the model. The specific sub-steps include: a personnel screening sub-step, a cognitive test sub-step, and a data entry sub-step. The personnel screening sub-step is used to screen out eligible personnel to participate in the cognitive test step. The cognitive test sub-step is used to conduct MemTrax tests on the screened eligible personnel. The data entry sub-step is used to enter the data. The data includes the basic information of the eligible personnel and the test data obtained in the cognitive test sub-step. The basic information includes age and education level. The test data includes reaction time and / or accuracy rate.

[0020] The data collation is used to eliminate the invalid data and ensure the validity of the data.

[0021] The data analysis constructs the model through the analysis of the data. The specific sub-steps include an educational level classification sub-step and a model result sub-step. The educational level classification sub-step is used to classify the data according to the educational level in the basic information, and obtain educational classification sub-data with the same number of classifications as the educational level. The model result sub-step is used to analyze and model each piece of the educational classification sub-data to obtain the age-response time correspondence relationship between the age and the response time, and / or the age-accuracy rate correspondence relationship between the age and the accuracy rate under each category of educational level.

[0022] In some embodiments, a method for constructing a model for large-scale screening of cognitive impairment based on MemTrax provided by the present invention:

[0023] According to the model construction method described in claim 1, it is characterized in that:

[0024] The model result sub-step further includes obtaining the age-comprehensive score correspondence relationship between the age and the comprehensive score under each category of educational level. The comprehensive score is a score given based on the comprehensive performance combining the response time and the accuracy rate.

[0025] In some embodiments, a method for constructing a model for large-scale screening of cognitive impairment based on MemTrax provided by the present invention:

[0026] The comprehensive score = accuracy rate / response time * n, where n ≠ 0.

[0027] Or, the comprehensive score = a * accuracy rate - b * response time, where a ≠ 0 and b ≠ 0.

[0028] In some embodiments, a method for constructing a model for large-scale screening of cognitive impairment based on MemTrax provided by the present invention:

[0029] The comprehensive score is a variable obtained by reducing the dimensions of the two variables of the accuracy rate and the response time through principal component analysis.

[0030] In some embodiments, a method for constructing a model for large-scale screening of cognitive impairment based on MemTrax provided by the present invention:

[0031] The data analysis is implemented based on the GAMLSS modeling method: under each educational level, the age-response time correspondence relationship, and / or the age-accuracy rate correspondence relationship, and / or the age-comprehensive score correspondence relationship are respectively established.

[0032] In a second aspect, an apparatus for large-scale screening of cognitive impairment applying the model construction method as described in the first aspect according to an embodiment of the present invention includes:

[0033] The MemTrax test module is used to perform the MemTrax test on the object to be evaluated and obtain the results of the MemTrax test.

[0034] The correspondence display module is used to display the age - reaction time correspondence and / or the age - correct rate correspondence and / or the age - comprehensive score correspondence.

[0035] According to the educational level and age of the object to be evaluated and the results of the MemTrax test obtained through the MemTrax test module, search in the age - reaction time correspondence and / or the age - correct rate correspondence and / or the age - comprehensive score correspondence displayed by the correspondence display module to determine the position where the cognitive ability level of the object to be evaluated is located, that is, the cognitive percentile of the object to be evaluated.

[0036] Confirm the risk level according to the cognitive percentile of the object to be evaluated, and issue a corresponding risk prompt according to the risk level.

[0037] In some embodiments, the cognitive impairment large - scale screening device applying the model construction method as described in the first aspect of the present invention further includes:

[0038] The weighing module is used to confirm the risk level according to the age - comprehensive score correspondence when the risk levels confirmed according to the reaction time and the correct rate are inconsistent.

[0039] In some embodiments, the cognitive impairment large - scale screening device applying the model construction method as described in the first aspect of the present invention:

[0040] The risk level is divided into high risk and extremely high risk.

[0041] When the cognitive percentile of the object to be evaluated is lower than the lower limit 2 but higher than the lower limit 1, it is the high risk, and a high - risk prompt is issued. When the cognitive percentile of the object to be evaluated is lower than the lower limit 1, it is the extremely high risk, and an extremely high - risk prompt is issued. The lower limit 1 and the lower limit 2 are ranges selected conventionally when establishing the reference value.

[0042] In some embodiments, the cognitive impairment large - scale screening device applying the model construction method as described in the first aspect of the present invention further includes:

[0043] The evaluation module is used to find out the results of the MemTrax test with statistical differences and confirm the corresponding cognitive percentile of the object to be evaluated. The evaluation module performs the following processing:

[0044] Taking a certain number of objects to be evaluated with common characteristics as the evaluation group,

[0045] Use one of the reaction time and the accuracy rate as the standard value, and the other as the comparison value.

[0046] Select an interval of the standard value, classify the interval at a specified interval, and determine whether there is a statistical difference between the value of the comparison value corresponding to each category in the evaluation group and the value in the eligible personnel. When there is such a statistical difference, recommend the preferred parameter or confirm the cognitive percentile of the evaluation object in the age-comprehensive score correspondence. The preferred parameter refers to one or two of the most suitable ones among the age-reaction time correspondence, the age-accuracy rate correspondence, and the age-comprehensive score correspondence.

[0047] In some embodiments, the cognitive impairment large-scale screening device applying the model construction method as described in the first aspect of the present invention further includes:

[0048] An evaluation module for giving a basis for confirming the cognitive percentile of the evaluation object according to the result of the MemTrax test:

[0049] For the evaluation object with the accuracy rate of 60%-70%, it is the age-comprehensive score correspondence or the age-reaction time correspondence.

[0050] For the evaluation object with the accuracy rate of 72%-94%, it is the age-comprehensive score correspondence.

[0051] For the evaluation object with the accuracy rate of 96%-100%, it is the age-comprehensive score correspondence or the age-reaction time correspondence.

[0052] The beneficial effects of the present invention are:

[0053] 1. The present invention simultaneously reflects the two important factors of education and age in the model, and in the way of classifying by education level and obtaining the corresponding relationship between age and MemTrax test data, reflects the relationship between the two important factors of education and age and cognitive impairment, which is an important innovation point of the present invention. This innovation point not only has substantial characteristics in the implementation of the technical solution, but also makes remarkable progress in the effect. By simultaneously observing the two factors of age and education, the present invention discovers evidence of the interaction between age and education level, and in the application, the influences of the two important factors can also be separated, providing a more objective reference standard and basis for the identification of cognitive impairment, more effectively distinguishing the risk of early cognitive impairment, and being beneficial to realizing early diagnosis and treatment of diseases and advancing the intervention window.

[0054] 2. The present invention introduces the concept of comprehensive score, balances the bias between the correct rate and the reaction time, making the test results more comprehensive, more robust, and thus more valuable as a reference.

[0055] 3. It helps to identify high-risk groups and diseased populations with cognitive impairment for early identification of cognitive impairment. There is almost no learning effect, which is suitable for observing the long-term development of individual cognition, identifying physiological and pathological declines in cognitive function, facilitating early intervention, and can be used as a tool for observing the efficacy of cognitive intervention research.

[0056] 4. This study only needs to collect age and education level and conduct computerized MemTrax standardized assessment. The objectivity of the indicators is high. There is no need to evaluate the consistency of the examiners. There is almost no learning effect. The picture recognition memory used is less affected by cultural background, is simple and economical, and is suitable for large-scale cognitive screening and triage management in physical examinations, communities, etc. It can also be used as one of the efficacy indicators for cognitive intervention follow-up. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] To more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. However, those skilled in the art should understand that the drawings described below are only some examples of the present application and do not limit its scope.

[0058] Figure 1 is a flowchart of a specific embodiment of the construction and use of the MemTrax-based large-scale screening model for cognitive impairment of the present invention.

[0059] Figure 2 is a flowchart of data arrangement of a specific embodiment of the construction method of the MemTrax-based large-scale screening model for cognitive impairment of the present invention.

[0060] Figure 3 is a detrended Q-Q plot of a specific embodiment of the construction method of the MemTrax-based large-scale screening model for cognitive impairment of the present invention.

[0061] Figure 4 is an age-based percentile curve graph of MTx-RT, MTx-%C, and MTx-Cp for different education categories of a specific embodiment of the construction method of the MemTrax-based large-scale screening model for cognitive impairment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0062] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Those skilled in the art should understand that the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments in the present application, those skilled in the art can make any appropriate modifications or variations to obtain all other embodiments.

[0063] In a first aspect, an embodiment of the present invention provides a method for constructing a model for large-scale screening of cognitive impairment based on MemTrax. The method for constructing the model includes the following steps:

[0064] Data collection, data sorting, and data analysis.

[0065] The data collection is used to collect the data required for constructing the model. The specific sub-steps include: a personnel screening sub-step, a cognitive test sub-step, and a data entry sub-step. The personnel screening sub-step is used to screen out eligible personnel to participate in the cognitive test step. The cognitive test sub-step is used to conduct a MemTrax test on the screened eligible personnel. The data entry sub-step is used to enter the data. The data includes the basic information of the eligible personnel and the test data obtained in the cognitive test sub-step. The basic information includes age and education level. The test data includes reaction time and / or accuracy rate.

[0066] The data sorting is used to eliminate the invalid data and ensure the validity of the data.

[0067] The data analysis is used to construct the model through the analysis of the data. The specific sub-steps include an education level classification sub-step and a model result sub-step. The education level classification sub-step is used to classify the data according to the education level in the basic information to obtain education classification sub-data with the same number of classifications as the education level. The model result sub-step is used to analyze and model each piece of education classification sub-data to obtain the age-reaction time correspondence relationship between the age and the reaction time, and / or the age-accuracy rate correspondence relationship between the age and the accuracy rate under each category of education level.

[0068] In this embodiment, the method for constructing a model for large-scale screening of cognitive impairment based on MemTrax includes the data collection, the data sorting, and the data analysis.

[0069] This data collection provides the data required by the model. To provide qualified data as much as possible, the participants need to be screened. If the personnel are not properly screened, unqualified data may be introduced, affecting the accuracy of the model. For the screening of personnel, standards suitable for the application of the established model should be formulated. The eligible personnel selected need to undergo the MemTrax test. The requirements for the MemTrax test conducted for data collection to establish the model should also be higher than those of ordinary MemTrax tests to collect more accurate data. The data entered includes the basic information of the eligible personnel and the test data obtained in the cognitive test sub-step. The test data can be the value directly returned after local testing or the return value obtained through online testing. The basic information includes age and education level. Of course, it can also include other information such as gender and medical history. The test data includes reaction time and / or accuracy rate. It should be emphasized here that in the process of data collection, both age and education level are included, and both of these variables are considered in the modeling, and a model covering both of these factors is established, which is an important invention point of the present invention. Specifically, it will be elaborated later. Here, it should be noted that the personnel screening sub-step, the cognitive test sub-step, and the data entry sub-step are mutually intersecting. For example, in the R & D process of the present invention, a total of 29,379 complete MemTrax tests were conducted. Inevitably, some people were undergoing screening, some were undergoing testing, and some were entering data. In addition, the data entry sub-step itself is also integrated with the personnel screening sub-step and the cognitive test sub-step. Relevant personnel can undergo a screening while entering their own basic information, or enter the basic information synchronously during the screening process, which is all possible. The current MemTrax test can be conveniently completed through electronic devices. Therefore, during the test, the test data can be synchronously transmitted for data entry work.

[0070] This data sorting, also known as data cleaning, is used to eliminate invalid data and ensure the validity of the data. Invalid data is data that is not beneficial for establishing the model and cannot achieve the purpose of establishing the model. For example, data with careless testing, repeated testing, no response, invalid testing, and abnormal results.

[0071] This data analysis constructs the model through the analysis of the data. The specific sub-steps include the educational level classification sub-step and the model result sub-step. The educational level classification sub-step and the model result sub-step are not completely separate and isolated steps, but are combined according to the needs of the model used during the modeling process. It is emphasized here that during the modeling process, the important dimension of education should be considered in a classified manner. In previous models based on MemTrax, the important factor of education was not included. At the same time, the two important factors of education and age are reflected in the model, and in the way that after classifying by educational level, the corresponding relationship between age and MemTrax test data is obtained, the relationship between these two important factors of education and age and cognitive impairment is reflected, which is an important innovation point of the present invention. This innovation point not only has substantial characteristics in the implementation of the technical solution, but also has made remarkable progress in terms of effects. The present invention discovers evidence of the interaction between age and educational level through the simultaneous observation of these two factors of age and education, and moreover, in the application, the influences of these two important factors can also be separated, providing a more objective reference standard and basis for the identification of cognitive impairment, more effectively distinguishing the risk of early cognitive impairment, which is beneficial to achieving early diagnosis and treatment of the disease and advancing the intervention window. The results of the corresponding relationships obtained in the model result sub-step are presented in the form of percentiles, so as to facilitate the participants in the MemTrax test to locate their respective levels in the population. The form of percentiles includes percentile curves and percentile tables, so as to facilitate people to accurately find their corresponding levels according to the test results in subsequent applications.

[0072] The model can be established by various mathematical methods, such as using the deviation method, percentile method, regression method, LMS method, etc., all of which can achieve the functions of this implementation plan. During the research process of the present invention, the GAMLSS modeling method is used. When establishing the model, not only three parameters such as median, standard deviation, and skewness are included, but also the kurtosis parameter is included. The percentile curves or tables formulated by this GAMLSS modeling method can better reflect the authenticity of the original data, and the modeling effect is better than the standards formulated by the aforementioned methods.

[0073] Based on this embodiment, preferably, the educational level is divided into three categories: high school and below is low educational level; junior college is secondary educational level; undergraduate and above is high educational level. The quantity and method of the aforementioned classification of the educational level are relatively balanced choices made through the analysis of data and comprehensive consideration of the convenience and effectiveness of use during the research process of the present invention.

[0074] In some embodiments, the present invention provides a method for constructing a model for large-scale screening of cognitive impairment based on MemTrax:

[0075] The model construction method according to claim 1 is characterized in that:

[0076] This model result sub-step further includes obtaining the age-comprehensive score correspondence relationship between the age and the comprehensive score at each education level of each category. The comprehensive score is a score given based on the comprehensive performance combining the reaction time and the correct rate.

[0077] In this embodiment, the age-comprehensive score correspondence relationship between the age and the comprehensive score included in this model result sub-step is an important innovation point of the present invention. In the prior art, models based on MemTrax all have a trade-off between the correct rate and the reaction time, usually building a model for each of these two indicators separately. The advantage of the prior art is that the operation is relatively simple and convenient, but the disadvantage is also obvious, that is, the results of the two models for the correct rate and the reaction time respectively are still biased, which is likely to cause confusion to users. In particular, during the research process of the present invention, it is found that users who participate in the test online by themselves tend to overly pursue the correct rate at the expense of the reaction time, which greatly reduces the reference value of the test results. Therefore, the present invention introduces the concept of the comprehensive score, balances the bias between the correct rate and the reaction time, makes the test results more comprehensive, more robust, and thus more valuable for reference. The comprehensive score is a score given based on the comprehensive performance combining the reaction time and the correct rate. The comprehensive score used by the present invention during the R & D process is MTx-Cp = correct rate / reaction time * 100. Of course, similar deformations can also be used, or a comprehensive score that can synthesize the reaction time and the correct rate can be constructed separately.

[0078] In some embodiments, a model construction method for large-scale screening of cognitive impairment based on MemTrax provided by the present invention:

[0079] The comprehensive score = correct rate / reaction time * n, n ≠ 0.

[0080] Or, the comprehensive score = a * correct rate - b * reaction time, a ≠ 0 and b ≠ 0.

[0081] In this embodiment, two ways of constructing the comprehensive score are given: 1) MTx-Cp = correct rate / reaction time * n (n can be any number other than 0); 2) MTx-Cp = a * correct rate - b * reaction time (a and b can be any number other than 0). These two constructions can both synthesize the effects of the correct rate and the reaction time, and give a more balanced and definite conclusion when the conclusions given based on the correct rate and the reaction time respectively conflict.

[0082] In some embodiments, a model construction method for large-scale screening of cognitive impairment based on MemTrax provided by the present invention:

[0083] The comprehensive score is a variable obtained by reducing the dimensionality of the two variables of the accuracy rate and the reaction time through principal component analysis.

[0084] In this embodiment, another way to construct the comprehensive score is provided, that is, by reducing the dimensionality of the two variables of the accuracy rate and the reaction time into one variable through principal component analysis, etc., to achieve the comprehensive evaluation of the accuracy rate and the reaction time.

[0085] In some embodiments, a method for constructing a model for large-scale screening of cognitive impairment based on MemTrax provided by the present invention:

[0086] The data analysis is implemented based on GAMLSS (Generalized Additive Models for Location, Scale and Shape): under each educational level, the age-reaction time correspondence, and / or the age-accuracy rate correspondence, and / or the age-comprehensive score correspondence are established respectively.

[0087] In this embodiment, in the cognitive field, GAMLSS is creatively introduced for the first time, presenting the changing trend of cognitive function with age in a non-linear manner, and supporting parameter tuning to select the best model. For example, methods such as residual statistics, Q-test, and detrended Q-Q plot method are used to adjust parameters, and the Q-test and detrended Q-Q plot method are used to evaluate the fitting effect. This modeling method not only incorporates three parameters such as median, standard deviation, and skewness, but also incorporates the kurtosis parameter. The formulated percentile curves or tables can better reflect the authenticity of the original data, which is superior to methods such as the deviation method, percentile method, regression method, and LMS method. The following takes the code for establishing age-specific cognitive percentile curves for subjects with different educational levels through the GAMLSS modeling method in R language as an example for detailed description.

[0088] Reaction time - High education

[0089] MTx-RT high education = gamlss(resp_time ~ cs(age, df = 3), sigma.formula = ~ cs(age, 4),

[0090] nu.formula = ~ 1, tau.formula = ~ 1, family = BCPE,

[0091] data = na.omit(data_MTx-RT high education))

[0092] Reaction time - Medium education

[0093] MTx-RT with secondary education = gamlss(resp_time ~ cs(age, df = 3), sigma.formula = ~ cs(age, 3),

[0094] nu.formula = ~ cs(age, df = 1), tau.formula = ~ 1, family = BCPE,

[0095] data = na.omit(data_MTx-RT with secondary education))

[0096] Response time - low education

[0097] MTx-RT with low education = gamlss(resp_time ~ cs(age, df = 3), sigma.formula = ~ cs(age, 3),

[0098] family = BCT, nu.formula = ~ 1, tau.formula = ~ 1,

[0099] data = na.omit(data_MTx-RT with low education))

[0100] Accuracy rate - high education

[0101] MTx-%C with high education = gamlss(correct_per ~ cs(age, df = 5), sigma.formula = ~ cs(age, 4),

[0102] nu.formula = ~ 1, tau.formula = ~ 1, family = BCPE,

[0103] data = na.omit(data_MTx-%C with high education))

[0104] Accuracy rate - secondary education

[0105] MTx-%C with secondary education = gamlss(correct_per ~ cs(age, df = 3), sigma.formula = ~ cs(age, 4),

[0106] nu.formula = ~ 1, tau.formula = ~ 1, family = BCPE,

[0107] data = na.omit(data_MTx-%C with secondary education))

[0108] Accuracy rate - low education

[0109] MTx-%C lower education level = gamlss(correct_per ~ cs(age, df = 4), sigma.formula = ~ cs(age, 3),

[0110] family = BCPE, nu.formula = ~ 1, tau.formula = ~ cs(age, 3),

[0111] data = na.omit(data_MTx-%C lower education))

[0112] Comprehensive index - higher education level

[0113] MTx-Cp higher education level = gamlss(rate ~ cs(age, df = 3), sigma.formula = ~ cs(age, 4),

[0114] nu.formula = ~ 1, tau.formula = ~ 1, family = BCPE,

[0115] data = data_MTx-Cp higher education)

[0116] Comprehensive index - middle education level

[0117] MTx-Cp middle education level = gamlss(rate ~ cs(age, df = 4), sigma.formula = ~ cs(age, 3),

[0118] nu.formula = ~ 1, tau.formula = ~ 1, family = BCPE,

[0119] data = data_MTx-Cp middle education)

[0120] Comprehensive index - lower education level

[0121] MTx-Cp lower education level = gamlss(rate ~ cs(age, df = 3), sigma.formula = ~ cs(age, 3),

[0122] family = BCPE, nu.formula = ~ 1, tau.formula = ~ 1,

[0123] data = na.omit(data_MTx-Cp lower education))

[0124] Second, an apparatus for large-scale screening of cognitive impairment applying the model construction method as described in the first aspect is proposed in an embodiment of the present invention. The apparatus for large-scale screening of cognitive impairment includes:

[0125] The MemTrax test module is used to conduct the MemTrax test on the evaluation object to obtain the result of the MemTrax test.

[0126] The corresponding relationship display module is used to display the age - reaction time corresponding relationship and / or the age - correct rate corresponding relationship and / or the age - comprehensive score corresponding relationship.

[0127] According to the educational level and age of the evaluation object and the result of the MemTrax test obtained through the MemTrax test module, search in the age - reaction time corresponding relationship and / or the age - correct rate corresponding relationship and / or the age - comprehensive score corresponding relationship displayed by the corresponding relationship display module to determine the position where the cognitive ability level of the evaluation object is located, that is, the cognitive percentile of the evaluation object.

[0128] Confirm the risk level according to the cognitive percentile of the evaluation object, and issue a corresponding risk prompt according to the risk level.

[0129] In this embodiment, a large - scale screening device for cognitive impairment applying the model constructed by the model construction method according to the first aspect is given. The large - scale screening device for cognitive impairment includes the MemTrax test module and the corresponding relationship display module, which can help complete the large - scale screening of cognitive impairment. In the large - scale screening, the personnel participating in the MemTrax test are the evaluation objects, and there are no requirements and restrictions on the evaluation objects, as long as they can complete the MemTrax test. After the evaluation object undergoes the MemTrax test and obtains the test result, then according to their own educational level, age and the result of the MemTrax test, and based on the result displayed by the model obtained by the model construction method according to the first aspect, search for the position where their cognitive ability level is located in the age - reaction time corresponding relationship and / or the age - correct rate corresponding relationship and / or the age - comprehensive score corresponding relationship, that is, the cognitive percentile of the evaluation object, and then confirm the risk level where their cognitive state is located accordingly. The risk level is divided according to whether there is a potential risk of suffering from cognitive impairment. If there is a potential risk, early prevention and early intervention should be carried out to control the risk in time to prevent the occurrence of cognitive impairment, or to control the condition in time after the occurrence of cognitive impairment to prevent further development. The division of the risk level should select conventional and generally recognized standards in combination with the progress of medical research, and may vary according to different regions, living habits, etc., and can be adjusted as needed.

[0130] It should be supplemented and explained here that during the continuous monitoring of the cognitive level, when the cognitive percentile of the evaluation object drops sharply within a short period of time, even if the foregoing standards are not met, further clinical cognitive assessment is still recommended.

[0131] In some embodiments, the cognitive impairment large-scale screening device applying the model building method as described in the first aspect of the present invention further includes:

[0132] A weighing module, configured to, when the risk levels respectively confirmed according to the reaction time and the correct rate are inconsistent, confirm the risk level according to the age-comprehensive score correspondence.

[0133] In this embodiment, a solution is given when the results of the correct rate and the reaction time conflict, that is, through the weighing module, the age-comprehensive score correspondence is used to confirm the cognitive percentile of the evaluation object. There will be a weighing phenomenon during the MemTrax test, such as sacrificing the reaction time in pursuit of the correct rate, or sacrificing the correct rate in pursuit of the reaction time. In such a situation, the existing methods of respectively using the correct rate or the reaction time cannot accurately reflect the cognitive status, but the comprehensive score of this embodiment can accurately reflect the cognitive status.

[0134] In some embodiments, the cognitive impairment large-scale screening device applying the model building method as described in the first aspect of the present invention:

[0135] The risk level is divided into high risk and extremely high risk.

[0136] When the cognitive percentile of the evaluation object is lower than the lower limit 2 but higher than the lower limit 1, it is the high risk, and a high-risk prompt is issued. When the cognitive percentile of the evaluation object is lower than the lower limit 1, it is the extremely high risk, and an extremely high-risk prompt is issued. The lower limit 1 and the lower limit 2 are ranges conventionally selected when establishing the reference value.

[0137] In this embodiment, preferably, a method for dividing the risk levels is given. The risk levels are divided into high risk and extremely high risk according to the lower limit 2 and the lower limit 1. For example, when the cognitive percentile of the evaluation object is between the lower limit 1 and the lower limit 2, it belongs to high risk, indicating that the person may be at high risk of cognitive impairment; when the cognitive percentile of the evaluation object is lower than the lower limit 1, it belongs to extremely high risk, indicating that there may be cognitive abnormalities. The lower limit 1 and the lower limit 2 should be selected as conventional and generally recognized standards according to the progress of medical research, and may vary according to different regions, living habits, etc., and can be adjusted as needed. The lower limit 1 is the risk critical point between having cognitive impairment and not having it, that is, when it is lower than the lower limit 1, according to the consensus in the medical field, there is a relatively high probability of having a disease related to cognitive impairment. The lower limit 2 is the demarcation point for the potential risk of suffering from cognitive impairment diseases, that is, when it is higher than the lower limit 1 but lower than the lower limit 2, according to the consensus in the medical field, the probability of suffering from cognitive impairment diseases in the future is relatively high. According to the research of the present invention, currently, the recommended values of the lower limit 1 and the lower limit 2 are 3% and 10% respectively. In addition, it should be supplemented that when the cognitive percentile of the evaluation object is higher than the lower limit 2, it is determined as low risk, and the risk prompt does not need to be issued. Of course, it is also possible to inform the evaluation object.

[0138] In some embodiments, the cognitive impairment large-scale screening device applying the model construction method as described in the first aspect provided by the present invention further includes:

[0139] An evaluation module, configured to find out the results of the MemTrax test with statistical differences and confirm the corresponding cognitive percentile of the evaluation object. The evaluation module performs the following processing:

[0140] Taking a certain number of evaluation objects with common characteristics as an evaluation group,

[0141] Taking one of the reaction time and the correct rate as a standard value and the other as a comparison value.

[0142] Selecting an interval of the standard value, classifying the interval at a specified interval, and determining whether there are statistical differences between the values of the comparison value corresponding to each category in the evaluation group and the values in the eligible personnel. When there is such a statistical difference, the preferred parameter is recommended or the cognitive percentile of the evaluation object is confirmed in the age-comprehensive score correspondence. The preferred parameter refers to one or two of the most suitable ones in the age-reaction time correspondence, the age-correct rate correspondence, and the age-comprehensive score correspondence.

[0143] In this embodiment, through the evaluation module, a specific quantitative method for more scientifically conducting cognitive evaluation is given, which indicates which indicators should be preferentially considered under what circumstances. The evaluation objects in the evaluation group all have common characteristics. For example, they all participate in the evaluation online, or they all come from a certain region, community, etc., or they all suffer from a certain disease. Therefore, they may have some common tendencies with statistical significance, that is, they have a significant speed-accuracy trade-off, sacrificing reaction time in pursuit of high accuracy, or sacrificing accuracy in pursuit of reaction time. In order to accurately point out which indicators should be preferentially considered under what circumstances, in this embodiment, one of the reaction time and the accuracy rate is used as the standard value, and the other is used as the comparison value. For the convenience of narration and to make the application of this embodiment clearer to the public, more specific content is substituted below. Suppose the accuracy rate is used as the standard value, then the reaction time is the comparison value. For example, in the MemTrax test, there are 50 true-or-false questions, and 0-100 represents 0%-100%. Then the accuracy rate is an even number between 0 and 100. Generally, the accuracy rate of the evaluation object is not lower than 60 (in the research of the present invention, the test with MTx-%C < 60% is regarded as an invalid evaluation). Therefore, for the sake of simplicity in calculation and efficiency improvement here, the interval of 60-100 for the accuracy rate can be selected for comparison. The values in this interval are classified at an interval of 2, the minimum interval in the specific scheme, and it is successively determined whether there is a statistically significant difference between the value of the comparison value in the evaluation group, that is, the value of the reaction time, and the value of the comparison value in the eligible personnel, that is, the value of the reaction time, when the accuracy rate is 60, 62, 64... 100. When there is a statistical difference, it indicates that the evaluation group has a significant speed-accuracy trade-off. At this time, the age-comprehensive score correspondence relationship should be used to confirm the cognitive percentile of the evaluation object, or the cognitive percentile of the evaluation object should be confirmed according to the recommended preferred parameter. Specifically, in the research process of the present invention, it is found that among the people who conduct the MemTrax test online, when MTx-%C is between 72 and 94, the online MTx-RT is significantly longer than the offline MTx-RT (P < 0.05). Therefore, there is a reaction time-accuracy rate trade-off, that is, in the online test, more tendency is to sacrifice reaction time to pursue higher accuracy. Here, P is a parameter for statistically determining hypothesis testing. When P > 0.05, it indicates no statistical difference; when P < 0.05, it indicates a statistical difference. At this time, the age-comprehensive score correspondence relationship or the recommended preferred parameter should be used to confirm the cognitive percentile of the evaluation object.

[0144] In some embodiments, the cognitive disorder large-scale screening device applying the model construction method as described in the first aspect provided by the present invention further includes:

[0145] An evaluation module, which is used to give a basis for confirming the cognitive percentile of the evaluated object according to the results of the MemTrax test:

[0146] For the evaluated object with a correct rate of 60%-70%, it is the age-comprehensive score correspondence or the age-response time correspondence.

[0147] For the evaluated object with a correct rate of 72%-94%, it is the age-comprehensive score correspondence.

[0148] For the evaluated object with a correct rate of 96%-100%, it is the age-comprehensive score correspondence or the age-response time correspondence.

[0149] In this embodiment, through the evaluation module, the order of preferentially recommending which indicators to use in various situations is directly given, which is conducive to a more scientific specific quantification method for cognitive evaluation. This recommended order is obtained after analyzing the trade-off phenomenon of online and offline test results. In addition, the fitting effect of each model and the existence of the ceiling effect are combined. When the evaluated objects are all measured online, since online evaluation is a relatively common method, and the range of the trade-off phenomenon may not only apply to online measurement, but has a certain universality. Therefore, preferably, the recommended order under this standard is recommended to simplify the workload in large-scale screening. After all, the purpose of the present invention is to provide certain reference for doctors' diagnosis, rather than replacing doctors' diagnosis. Therefore, when weighing between convenience and accuracy, it can be tilted towards convenience.

[0150] Here, it should be supplemented that the technical solution of this embodiment can be further refined as:

[0151] For the evaluated object with a correct rate of 60%-70%, the age-comprehensive score correspondence or the age-response time correspondence is preferentially recommended, and the age-correct rate correspondence is the second.

[0152] For the evaluated object with a correct rate of 72%-94%, the age-comprehensive score correspondence is preferentially recommended, the age-response time correspondence is the second, and the age-correct rate correspondence is the third.

[0153] For the evaluated object with a correct rate of 96%-100%, the age-comprehensive score correspondence or the age-response time correspondence is preferentially recommended, and the age-correct rate correspondence is the second.

[0154] The following uses a real case in the R & D process of the present invention as an example to further elaborate on the technical solutions and beneficial effects proposed by the present invention. It should be understood that the detailed steps and parameters provided in the following embodiments are only for the elaboration and reference of the technical solutions of the present invention, and do not limit the technical solutions of the present invention. The following embodiments only select one of many technical paths for implementation, but do not represent that the technical solutions of the present invention can only be implemented in this way.

[0155] As Figure 1 shown, integrating the foregoing embodiments, it is a flowchart of the large-scale screening technology for cognitive impairment based on MemTrax. This flowchart successively shows the processes of constructing and using this model. The use includes, in addition to the evaluation for individuals, the feedback on the statistically significant differences obtained for groups. When constructing a model for large-scale screening of cognitive impairment based on MemTrax, personnel need to be screened first. As mentioned before, in order to provide qualified data as much as possible, the participants need to be screened. For the screening of personnel, standards adapted to the application of the constructed model should be formulated. Here, a specific example of such a standard is given for reference. The screening criteria include inclusion criteria and exclusion criteria. The inclusion criteria are: (1) aged 25 - 75 years old; (2) having a physical examination that meets the requirements; (3) evaluating MemTrax for the first time; (4) informed consent. The exclusion criteria are: (1) under 25 years old or over 75 years old; (2) being diagnosed with Alzheimer's disease or other cognitive impairment diseases; (3) physical impairments that prevent accurate completion of the MemTrax test.

[0156] For the eligible personnel screened out, the MemTrax test (evaluation) needs to be carried out. As mentioned before, the requirements for the MemTrax test carried out to collect data for model establishment should also be higher than those for ordinary MemTrax tests. The requirements referred to here refer to the requirements for auxiliary work before, during, and after the MemTrax test. Here, in order to help the public understand how to obtain a better-performing model, the requirements used in the R & D process of the present invention are provided for reference:

[0157] (1) Face-to-face test.

[0158] (2) Uniform training of the testers.

[0159] (3) Ensure that the subjects (i.e., the eligible personnel screened out) fully understand the rules. The understanding of the rules includes:

[0160] ① Emphasize the key points before the test, such as the pictures being exactly the same rather than similar; the faster the computer space bar is clicked, the better.

[0161] ② Without affecting the subject's test, during the test, determine whether the subject really understands the rules or has a serious attitude and make a record; if the subject does not understand the rules or has a non-serious attitude, the subject has another 2 chances to retake the MemTrax test, and each time the MemTrax test uses new pictures.

[0162] ③ If the reaction time (MTx-RT) in the evaluation result is longer than 1.4 seconds or the correct rate (MTx-%C) is lower than 81%, the tester needs to confirm again whether the subject really understands the rules.

[0163] This data collation, also known as data cleaning, eliminates the invalid data, such as Figure 2 shown, for example, data of non-serious testing, repeated testing, no response, invalid testing, and abnormal results. During the research and development process of the present invention, some standards were also specifically formulated. Here, only the standards during the research and development process are provided for reference. No response: MTx-RT < 300ms or MTx-RT > 2900ms; invalid evaluation: MTx-%C < 60%; abnormal results: MTx-%C > (mean MTx-％C + 5SD) or MTx-%C < (mean MTx-％C - 5SD), MTx-RT > (mean MTx-RT + 5SD) or MTx-RT < (mean MTx-RT - 5SD), and this process is repeated 3 times. Among them, SD is the standard deviation.

[0164] It should be understood that the reference information provided above only provides a reference for the public to better implement the technical solution of the present invention, and does not limit the technical solution of the present invention. Using more relaxed requirements to further strengthen the test effect, or to increase convenience or reduce costs, does not affect the substantial implementation of the technical solution of the present invention.

[0165] Specifically, the actual data during the research and development process of the present invention is as follows:

[0166] There were a total of 29,379 complete tests. 296 subjects who did not test seriously and refused to retest were deleted, 2,320 repeated tests were deleted, 115 non-responses or test failures were deleted, and 15 outliers were deleted, and a total of 26,633 subjects were included. The subjects here are the eligible personnel screened out.

[0167] Among the subjects, there were 12,862 (48.29%) females and 13,771 (51.71%) males. There were 3,705 (13.91%) people with a high school education or below, 3,624 (13.61%) people with a junior college education, 13,540 (50.84%) people with a bachelor's degree, and 5,764 (21.64%) people with a master's degree or above. The average age of the subjects was 43.0 ± 12.1 years.

[0168] In the process of data analysis, that is, in the process of constructing the model through the analysis of the data, the GAMLSS modeling method is used in the research and development process of the present invention. GAMLSS is often used to construct percentile standard curves. When establishing a model using the GAMLSS modeling method, not only three parameters, namely the median, standard deviation, and skewness, are incorporated, but also the kurtosis parameter is included. Therefore, the percentile curves formulated by it can better reflect the true nature of the original data, and the modeling effect is better than the standards formulated by methods such as the deviation method, percentile method, regression method, and LMS (lambda-mu-sigma, skewness coefficient-median-coefficient of variation method). The present invention innovatively applies the GAMLSS modeling method in the field of cognitive evaluation to construct a percentile curve and / or percentile table for cognitive test results based on education and age. Compared with the prior art, this is a cognitive model that is neither a linear nor a quadratic curve, can better reflect the true nature of the data, and is simple and easy to use.

[0169] The process of establishing the model can be divided into two stages. That is, in the first step, the classification of educational attainment is generated through multiple-factor linear regression analysis. In the second step, age-specific cognitive percentile curves and age-specific cognitive percentile tables are established for subjects with different educational attainments through the GAMLSS modeling method. The following is an introduction respectively.

[0170] In the first step, the classification of educational attainment is generated through multiple-factor linear regression analysis. Statistical analysis is performed using SPSS (Statistical Product Service Solutions) 22.0. Taking MTx-RT, MTx-%C, and MTx-Cp as the dependent variables and gender, education (dummy variable), and age (dummy variable) as the independent variables, multiple-factor linear regression is performed respectively. Three dummy variables are generated for educational attainment, namely education1, education2, and education3. Five dummy variables are generated for age: age1, age2, age3, age4, and age5. The specific settings of the dummy variables are as follows:

[0171] Table 3 Settings of dummy variables for educational attainment

[0172]

[0173]

[0174] Taking the population with high school education or below as the reference, the regression coefficients of MTx-RT for junior college, undergraduate, master's and above subjects were -0.062, -0.081, -0.080 (all P<0.05), for MTx-%C, the regression coefficients were 2.0160, 3.363, 3.693 (all P<0.05), and for MTx-Cp, the regression coefficients were 6.729, 9.817, 10.01 (all P<0.05). There was no significant difference between undergraduate and master's and above education levels, but there were significant differences from the other two educational levels (all P<0.05). Therefore, education was further integrated into the following three categories: low education (high school and below), medium education (junior college), and high education (undergraduate and above). The number of subjects from low to high was 3705 (13.91%), 3624 (13.61%), and 19304 (72.4%) respectively.

[0175] Table 4 Age dummy variable settings

[0176] age1 age2 age3 age4 age5 [25,30) 0 0 0 0 0 [30.40) 0 0 0 0 1 [40.50) 0 0 0 1 0 [50.60) 0 0 1 0 0 [60.70) 0 1 0 0 0 [70.75] 1 0 0 0 0

[0177] The age dummy variables classify age at fixed intervals. As shown in Table 4, the population aged 25 - 75 years was divided into 6 categories.

[0178] In the second step, age-specific cognitive percentile curves and age-specific cognitive percentile tables were established for subjects with different educational levels through the GAMLSS modeling method. The present invention uses the GAMLSS package on R 3.6.3 software to implement the percentile curve, and the goodness-of-fit test is performed through the detrended Q-Q plot and the goodness-of-fit Q test. The final model is the best model selected based on the global deviation (GD), Akaike information criterion (AIC), Schwarz Bayesian criterion (SBC), as well as the model results and residual diagnosis and the visual evaluation of the percentiles themselves.

[0179] The GAMLSS modeling method is presented in a specific distribution form of D(μ, σ, ν, τ), and its formula is:

[0180] y~D(μ,σ,ντ)

[0181] g1(μ)=s1(u)=x1β1+s11(x11)+...+s1J1(x1J1)

[0182] g2(σ)=s2(u)=x2β2+s21(x21)+...+s2J2(x2J2)

[0183] g3(ν)=s3(u)=x3β3+s31(x31)+...+s3J3(x3J3)

[0184] g4(τ) = s4(u) = x4β4 + s41(x41) +... + s4J4(x4J4)

[0185] u = xξ

[0186] The D distribution represents sub - distribution models such as BCCG, BCPE, or BCT. Each sub - distribution model usually contains 4 parameters: The first parameter is a parameter reflecting the distribution position, such as the mean, median, etc.; the second parameter is a parameter reflecting the scale (i.e., data discreteness), such as the standard deviation, mean square deviation, coefficient of variation, etc.; the third and fourth parameters are parameters reflecting the shape of the distribution, such as skewness and kurtosis. In addition, the g(·) function in the above formula represents an appropriate link function. s(·) represents a non - parametric smoothing function and the hyperparameter ξ, where ξ is the power - transformation exponent of the explanatory variable x (xξ), and the power - transformation of x can expand the scale of x, thereby improving the fitting effect of the smooth curve. In GAMLSS, the default link function for μ in BCCGo, BCPEo, and BCTo is log, and the non - parametric smoothing function s(·) usually needs to be determined by the degrees of freedom df of each parameter.

[0187] During the research process, with age as the independent variable x, the response variables y are modeled for MTx - RT, MTx - %C, and MTx - Cp respectively. μ (median), σ (standard deviation), ν (skewness), and τ (kurtosis) are non - parametric functions of x. The response variables are no longer limited to the exponential distribution, and the best model can be selected by continuously adjusting the parameters. First, iterate through the sub - distribution models of GAMLSS, such as BCCG(μ, σ, ν, τ), BCT(μ, σ, ν, τ), and BCPE(μ, σ, ν, τ), etc. Given that the sample size of this study is n > 1000, according to the generalized minimum Akaike information criterion, the optimal model is selected based on the minimum value of SBC. Except that the optimal model for MTx - RT with low education level is the BCT distribution model, the optimal models for MTx - RT, MTx - %C, and MTx - Cp with high, medium, and low education levels are all BCPE distribution models.

[0188] To fit the parameter curves, this study used residual statistics, Q-test, and detrended Q-Q plot methods to adjust the parameters, and used Q-test and detrended Q-Q plot methods to evaluate the fitting effect. Finally, the mean, variance, skewness, and kurtosis of the residuals of the MTx-RT, MTx-%C, and MTx-Cp fitting models for high, medium, and low educational levels were very close to the standard values of 0, 1, 0, and 3. In the Q-test, the Z scores Z1, Z2, and Z3 in the Z scores represent the standardized residuals of the fitting curves of the parameter mean, standard deviation, and skewness, respectively. If the standardized residual |Z| < 2 (P > 0.05), it indicates that the parameter fitting curve has a good fit. In the present invention, among the 20 Z3 values of MTx-RT for highly educated people, 1 Z3 value is greater than 2. However, considering that its overall p-value is greater than 0.05, the skewness of the MTx-RT model for highly educated people has a good fit. The overall p-values of other models are all greater than 0.05, and if individual Z values are greater than 2, it can also be considered that the model has a good fit (Tables 5 to 13). Figure 3 The detrended Q-Q plots show that the age-specific percentile curves of MTx-RT and MTx-Cp for high, medium, and low educational levels have a good fit, and the fitting effect of MTx-%C improves with the decrease in educational level.

[0189] Table 5 Q-test Z values of MTx-RT for highly educated people

[0190]

[0191]

[0192] Note: Absolute values of Z1, Z2, and Z3 greater than 2 represent poor fits of the mean, standard deviation, and skewness, respectively

[0193] Table 6 Q-test Z values of MTx-RT for moderately educated people

[0194]

[0195] Note: Absolute values of Z1, Z2, and Z3 greater than 2 represent poor fits of the mean, standard deviation, and skewness, respectively

[0196] Table 7 Q-test Z values of MTx-RT for low-educated people

[0197]

[0198]

[0199] Note: Absolute values of Z1, Z2, and Z3 greater than 2 represent poor fits of the mean, standard deviation, and skewness, respectively

[0200] Table 8 Q-test Z values of MTx-%C for highly educated people

[0201]

[0202]

[0203] Note: Absolute values of Z1, Z2, and Z3 greater than 2 respectively represent poor fitting of mean, standard deviation, and skewness.

[0204] Q-test Z values of MTx-%C for individuals with secondary education level in Table 9

[0205]

[0206] Note: Absolute values of Z1, Z2, and Z3 greater than 2 respectively represent poor fitting of mean, standard deviation, and skewness.

[0207] Q-test Z values of MTx-%C for individuals with low education level in Table 10

[0208]

[0209]

[0210] Note: Absolute values of Z1, Z2, and Z3 greater than 2 respectively represent poor fitting of mean, standard deviation, and skewness.

[0211] Q-test Z values of MTx-Cp for individuals with high education level in Table 11

[0212]

[0213]

[0214] Note: Absolute values of Z1, Z2, and Z3 greater than 2 respectively represent poor fitting of mean, standard deviation, and skewness.

[0215] Q-test Z values of MTx-Cp for individuals with secondary education level in Table 12

[0216]

[0217] Note: Absolute values of Z1, Z2, and Z3 greater than 2 respectively represent poor fitting of mean, standard deviation, and skewness.

[0218] Q-test Z values of MTx-Cp for individuals with low education level in Table 13

[0219]

[0220] Note: Absolute values of Z1, Z2, and Z3 greater than 2 respectively represent poor fitting of mean, standard deviation, and skewness.

[0221] Based on the above adjusted parameters, age-based percentile curves of MTx-RT, MTx-%C, and MTx-Cp for different education categories are as follows Figure 4As shown, the reference standards for percentile curves are shown in Tables 14 to 16.

[0222] Table 14 The reference standard for the MTx-RT(s) percentile curve

[0223]

[0224] Table 15: The reference standard for the MTx-%C(%) percentile curve

[0225]

[0226] Table 16: The reference standard for the MTx-Cp( / s) percentile curve

[0227]

[0228] To test whether there is an interaction between educational level and age, the present invention uses MTx-%C, MTx-RT, and MTx-CP as dependent variables, and educational level, gender, age, and age * educational level as independent variables, and performs regression analysis using SPSS 22.0. The results show that for MTx-RT, when (high education * age) is used as a reference, β 中等教育*年龄 is 0.002 (P < 0.05), and β 低教育*年龄 is 0.004 (P < 0.05); in MTx-%C, β 中等学历*年龄 is -0.061 (P < 0.05), and β 低学历*年龄 is -0.101 (P < 0.05). These research results indicate that there is a significant interaction between educational level and age. For example, the lower the educational level, the faster the deterioration of episodic memory and executive function. Similarly, the cognitive percentile curve graph shows that the MTx-Cp of higher and secondary education also remains stable until the mid-40s, and then decreases with age. For subjects with a lower educational level, MTx-Cp begins to decline steadily from the age of 25, and the lower the educational level, the faster MTx-Cp declines. The regression analysis results show that β 高等教育*年龄 is used as a reference; the β 中等学历*年龄 and β 低学历*年龄 values are -0.174 (P < 0.05) and -0.268 (P < 0.05), respectively. It can be seen that it is precisely because the technical solution of the present invention considers the factors of both education and age simultaneously during modeling that it reveals that the correlation between this educational level and this age is very large, and it also provides a relatively clear reference basis. For people of different ages and different educational levels, the evaluation criteria and preventive attention requirements are different. Therefore, the technical solution proposed by the present invention has important value and significance for the identification and prevention of cognitive impairment.

[0229] The beneficial effects of the comprehensive score proposed by the technical solution of the present invention are also supported by data. During the research of the present invention, there were 8,113 subjects online, including 3,527 (43.47%) males and 4,586 (56.53%) females, with an average age of 39.73 ± 11.48 years. The numbers of people with a high school education or below, junior college, undergraduate, and master's degree or above were 1,253 (15.44%), 1,142 (14.08%), 3,253 (40.10%), and 2,465 (30.38%), respectively. There were statistically significant differences in gender, age, and education level between the two groups tested online and offline. The MTx-RT differences between online and offline at each MTx-%C level are shown in Table 13. When MTx-%C is between 62 and 70 or 96 and 98, there is no statistically significant difference in MTx-RT between online and offline tests when the MTx-%C scores are the same (P≥0.05). However, when MTx-%C is between 72 and 94, the online MTx-RT is significantly longer than the offline MTx-RT (P<0.05), so there is a reaction time-accuracy trade-off, that is, in the online test, more emphasis is placed on sacrificing reaction time to pursue higher accuracy. When MTx-%C reaches the maximum value of 100, the offline MTx-RT is larger (P<0.05). Therefore, it can be clearly seen that the current technical solution's trade-off between reaction time and accuracy is biased when referring to a single result, and it will cause confusion when the evaluation object combines the two results for comparison and the results are inconsistent, so it lacks reference value. The technical solution of the present invention that gives a comprehensive score for comprehensively evaluating the accuracy and reaction time creatively solves this problem and provides a solution during the construction of the model. Thus, when the evaluation object uses the model for cognitive impairment screening, it can conveniently obtain more objective and accurate results, which has great value and significance for improving the accuracy and reliability of screening results.

[0230] Table 17: Multivariate linear regression analysis of online and offline data

[0231]

[0232] Note: Adjusted for age, gender, and education level

[0233] The following gives an application case of the technical solution of the second aspect of the present invention for reference.

[0234] Li Si, with a junior high school education, 57 years old, the results of the MemTrax test are that MTx-%C is 82% and MTx-RT is 1.5 s.

[0235] MTx-Cp = MTx-%C / MTx-RT * 100 = 82% / 1.5 * 100 = 54.67

[0236] With a junior high school education, which belongs to the low education level, referring to the table corresponding to the low education level, the MTx-%C percentile is 43%, the MTx-RT percentile is 5%, and the MTx-Cp percentile is 7%. MTx-%C indicates a normal cognitive level, but the MTx-RT percentile indicates that the person is at high risk of cognitive impairment, and the conclusions of the two are in conflict. The MTx-Cp percentile is 7%, indicating that Li Si belongs to the high-risk group of cognitive impairment; because MTx-Cp is a comprehensive index, it avoids the trade-off between MTx-%C and MTx-RT. The final conclusion mainly refers to MTx-Cp, indicating that the person may be at high risk of cognitive impairment, and it is recommended to conduct regular cognitive examinations or close follow-up tests.

[0237] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations according to the concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field based on the concept of the present invention through logical analysis, reasoning, or limited experiments on the basis of the prior art should be within the protection scope determined by the claims.

Claims

1. A method for constructing a model for large-scale screening of cognitive impairment based on MemTrax, characterized in that, The model construction method includes the following steps: Data collection, data sorting, and data analysis; The data collection is used to collect the data required for constructing the model; Specific sub-steps include: Personnel screening sub-step, cognitive test sub-step, data entry sub-step; The personnel screening sub-step is used to screen out eligible personnel to participate in the cognitive test step; The cognitive test sub-step is used to conduct the MemTrax test on the screened eligible personnel; The data entry sub-step is used to enter the data; The data includes the basic information of the eligible personnel and the test data obtained in the cognitive test sub-step; The basic information includes age and education level; The test data includes reaction time and / or accuracy rate; The data sorting is used to eliminate the invalid data and ensure the validity of the data; The data analysis constructs the model through the analysis of the data. The specific sub-steps include the education level classification sub-step and the model result sub-step; The education level classification sub-step is used to classify the data according to the education level in the basic information to obtain education classification sub-data with the same number as the classification of the education level; The model result sub-step is used to analyze and model each of the education classification sub-data to obtain the age-reaction time correspondence between the age and the reaction time, and / or the age-accuracy rate correspondence between the age and the accuracy rate, under each category of the education level; The data analysis is implemented based on the GAMLSS modeling method: Under each education level, the age-reaction time correspondence, and / or the age-accuracy rate correspondence, and / or the age-comprehensive score correspondence are established respectively.

2. The model construction method according to claim 1, wherein: The model result sub-step further includes obtaining the age-comprehensive score correspondence between the age and the comprehensive score under each category of the education level; The comprehensive score is a score given based on the comprehensive performance combining the reaction time and the accuracy rate.

3. The model construction method according to claim 2, wherein: The comprehensive score = accuracy rate / reaction time * n, n ≠ 0; Or, the comprehensive score = a * accuracy rate - b * reaction time, a ≠ 0 and b ≠ 0.

4. The model construction method according to claim 2, wherein: The comprehensive score is a variable obtained by reducing the dimensions of the two variables of the accuracy rate and the reaction time through principal component analysis.

5. A cognitive impairment large-scale screening device applying the model construction method according to any one of claims 1-4, characterized in that, The large-scale screening device for cognitive impairment includes: The MemTrax test module is used to conduct the MemTrax test on the evaluation object to obtain the result of the MemTrax test; The correspondence display module is used to display the age-reaction time correspondence and / or the age-accuracy rate correspondence and / or the age-comprehensive score correspondence; Based on the educational level and age of the evaluation object and the results of the MemTrax test obtained through the MemTrax test module, search in the age-response time correspondence and / or the age-accuracy rate correspondence and / or the age-comprehensive score correspondence displayed by the corresponding relationship display module to determine the position where the cognitive ability level of the evaluation object is located, that is, the cognitive percentile of the evaluation object; Confirm the risk level according to the cognitive percentile of the evaluation object, and issue a corresponding risk prompt according to the risk level.

6. The large-scale screening device for cognitive impairment according to claim 5, wherein The large-scale screening device for cognitive impairment further includes: A weighing module, configured to, when the risk levels confirmed according to the response time and the accuracy rate are inconsistent, confirm the risk level according to the age-comprehensive score correspondence.

7. The large-scale screening device for cognitive impairment according to claim 5, wherein: The risk levels are divided into high risk and extremely high risk; When the cognitive percentile of the evaluation object is lower than the lower limit 2 but higher than the lower limit 1, it is the high risk, and a high-risk prompt is issued; When the cognitive percentile of the evaluation object is lower than the lower limit 1, it is the extremely high risk, and an extremely high-risk prompt is issued; The lower limit 1 and the lower limit 2 are ranges conventionally selected when establishing the reference value, and are 3% and 10% respectively.

8. The large-scale screening device for cognitive impairment according to claim 5, wherein The large-scale screening device for cognitive impairment further includes: An evaluation module, configured to find out the results of the MemTrax test with statistical differences and confirm the corresponding cognitive percentile of the evaluation object; The evaluation module performs the following processing: Taking a certain number of evaluation objects with common characteristics as the evaluation group, taking one of the response time and the accuracy rate as the standard value and the other as the comparison value; Select an interval of the standard value, classify the interval at a specified interval, and determine whether the numerical values of the comparison value corresponding to each category in the evaluation group and in the qualified personnel have statistical differences; When there are statistical differences, recommend the preferred parameter or confirm the cognitive percentile of the evaluation object in the age-comprehensive score correspondence; The preferred parameter refers to one or two of the most suitable ones among the age-response time correspondence, the age-accuracy rate correspondence, and the age-comprehensive score correspondence.

9. The large-scale screening device for cognitive impairment according to claim 5, wherein The large-scale screening device for cognitive impairment further includes: An evaluation module, configured to give a basis for confirming the cognitive percentile of the evaluation object according to the results of the MemTrax test: For the evaluation object with the accuracy rate between 60% and 70%, it is the age-comprehensive score correspondence or the age-response time correspondence; For the evaluation object with the accuracy rate between 72% and 94%, it is the age-comprehensive score correspondence; For the evaluation object with the accuracy rate between 96% and 100%, it is the age-comprehensive score correspondence or the age-response time correspondence.

Images