A rural elderly mild cognitive impairment risk prediction and stratified nursing management system and method

Abstract

The application discloses a rural elderly mild cognitive impairment risk prediction and stratified nursing management auxiliary system and method, and belongs to the technical field of intelligent nursing and mobile health auxiliary screening. The system comprises an application path selection module, a data acquisition module, a data preprocessing module, a nursing evaluation and scale automatic scoring module, a risk prediction module, a risk stratification module, an explanation output module, a nursing suggestion generation module and a mobile terminal display module. Through the rapid screening version, simple variables are collected; through the standard evaluation version, GDS-15, IADL, PSQI, SSRS and other scale items are embedded and automatically scored; a pre-trained machine learning model is called to output MCI risk prediction probability, low, medium and high risks are divided, and corresponding reevaluation frequency and stratified nursing management suggestions are generated. The application can display the results through a webpage calculator or a WeChat applet, is used for primary screening assistance and risk communication, and does not replace clinical diagnosis.

Description

Technical Field

[0002] This invention belongs to the fields of smart nursing, elderly cognitive health management, machine learning risk prediction, standardized scale automatic scoring, and mobile health-assisted screening. Specifically, it relates to an auxiliary system and method for predicting the risk of mild cognitive impairment, risk stratification, result interpretation, standardized nursing assessment automatic scoring, and generating stratified nursing management suggestions for rural elderly people. It can be displayed through a web calculator, WeChat mini-program, or other mobile terminals. Background Technology

[0003] Mild cognitive impairment is a transitional state between normal cognitive aging and dementia. Rural elderly populations are relatively vulnerable in identifying and continuously managing the risk of cognitive decline due to factors such as education level, access to medical resources, chronic disease management, social support, and health behaviors. In practice, primary care nurses often need to quickly identify key individuals requiring attention and implement differentiated follow-up and health management based on their risk levels.

[0004] Existing screening methods rely heavily on single cognitive scales or human experience judgment, making it difficult to comprehensively utilize multidimensional information such as functional status, psychological sleep, chronic disease burden, sensory function, and social behavior. Traditional screening results often only indicate whether cognitive impairment is suspected, lacking visualized risk probability output and low, medium, and high risk stratification. In primary care settings, there is a lack of follow-up frequency, health education, chronic disease management, sleep and mood management, functional maintenance, and referral recommendations that directly correspond to risk levels.

[0005] Furthermore, when using scales such as GDS-15, IADL, PSQI, and SSRS at the primary care level, nursing staff often need to ask questions item by item, perform manual conversions, or sum the results manually. This can easily lead to fragmented processes, time-consuming scoring, and human error. In particular, the PSQI involves multiple component score conversions, and the SSRS involves different dimensions such as objective support, subjective support, and support utilization. Without automatic scoring and result integration functions, the continuity and scalability of mass screening at the primary care level will be affected.

[0006] Machine learning prediction models can integrate multidimensional variables and output probabilistic risk results. However, if presented only as statistical indicators such as AUC, calibration curves, and decision curves, they are difficult for primary caregivers and families of the elderly to understand, limiting their application. Therefore, an auxiliary system is needed that can integrate multidimensional health variables, automatically score standardized scales, call machine learning models, output individual risk probabilities, perform risk stratification, and generate nursing management recommendations. This system should allow the model results to be visualized in the form of web pages or mini-programs, serving the cognitive health screening and stratified management of the elderly in rural primary care settings. Summary of the Invention

[0007] I. Purpose of the Invention The purpose of this invention is to provide an auxiliary system and method for predicting the risk of mild cognitive impairment and for stratified nursing management in rural grassroots settings. Through rapid screening and standard assessment pathways, multidimensional variable collection, standardized scale itemized entry and automatic scoring, data preprocessing, machine learning risk prediction, risk stratification, interpretation output, and nursing suggestion generation, it can achieve visualized auxiliary assessment and stratified management of the risk of mild cognitive impairment in rural elderly people. Technical solution

[0008] To achieve the above objectives, this invention provides an auxiliary system for risk prediction and stratified nursing management of mild cognitive impairment in rural elderly people, including a data acquisition module, a data preprocessing module, a nursing assessment and automatic scale scoring module, a risk prediction module, a risk stratification module, an interpretation output module, a nursing suggestion generation module, a mobile terminal display module, and an optional data security and log management module. Each module can be deployed on a local front-end, a mini-program, a server, or implemented through a combination of front-end and back-end model interfaces.

[0009] The data acquisition module is used to collect basic information, chronic diseases and health status, behavioral and social factors, and standardized scale items for rural elderly people. The basic information includes age, years of education, body mass index, and average monthly household income per capita; the chronic diseases and health status includes hypertension, diabetes, coronary heart disease or other cardiovascular diseases, the number of chronic diseases, hearing problems, and vision problems; the behavioral and social factors include regular exercise, frequency of social participation, frequency of cultural, recreational, or intellectually stimulating activities, smoking, and alcohol consumption.

[0010] The nursing assessment and automatic scoring module converts standardized scales such as GDS-15, IADL, PSQI, and SSRS into an itemized input interface. Primary care nurses can select or fill in items based on interview results. The system automatically performs reverse scoring, dimension conversion, and total score calculation according to pre-designed scoring rules, and automatically transmits the resulting scale scores to the risk prediction module. The automatic scoring includes, but is not limited to: summing the reverse scores for items 1, 5, 7, and 11 of the GDS-15; summing the IADL items to obtain the total score for instrumental activities of daily living; converting PSQI items into 7 component scores and calculating a total score from 0 to 21; and summing SSRS items according to rules such as objective support, subjective support, and support utilization.

[0011] The data preprocessing module is used to perform format validation, range validation, missing value indication, variable encoding, and model adaptation on the input variables and automatic scoring results. For continuous variables, the system performs numerical validity judgment and necessary standardization or model adaptation; for binary variables, the system encodes them as 0 or 1; for ordinal categorical variables, the system converts them into ordinal variables or dummy variables required by the model.

[0012] The risk prediction module is used to invoke a trained machine learning model to calculate the preprocessed variables and output the predicted probability of an individual's mild cognitive impairment risk. The machine learning model can be XGBoost, Random Forest, Elastic Network, Logistic Regression, or other predictive models capable of outputting probabilities. In a preferred embodiment, the system uses the XGBoost model as the primary risk prediction model and a logistic regression model or noctilinear plot as an auxiliary interpretive tool.

[0013] The risk stratification module is used to classify risks into low, medium, and high risks based on the probability values ​​output by the risk prediction module. The risk stratification rules may include a first threshold and a second threshold, where risks below the first threshold are defined as low risk, risks between the first and second thresholds are defined as medium risk, and risks above or equal to the second threshold are defined as high risk. In one implementation, the first threshold is 0.20 and the second threshold is 0.40, meaning a predicted probability less than 0.20 is low risk, 0.20 to 0.39 is medium risk, and a predicted probability greater than or equal to 0.40 is high risk.

[0014] The interpretation output module is used to enhance the understandability of risk outcomes. Based on the importance of model features, SHAP value, or preset interpretation rules, this module can output key factors associated with increased individual risk, such as decreased instrumental daily living activities, shorter years of education, abnormal body mass index, poor sleep quality, more pronounced depressive symptoms, heavier burden of chronic diseases, insufficient social support, or less social participation.

[0015] The nursing recommendation generation module generates nursing management recommendations of varying intensities based on risk stratification results and key risk factors. For low-risk individuals, the recommendations primarily focus on cognitive health education, lifestyle guidance, routine chronic disease management, cognitive stimulation activities, and regular follow-up assessments. For medium-risk individuals, in addition to routine management, enhanced follow-up is conducted, with attention paid to sleep management, mood screening, chronic disease control, functional status, and social support. High-risk individuals are included in the key management group, and further cognitive screening or specialist assessment is recommended. Individualized nursing plans should be developed in conjunction with family members, and referral to neurology, geriatrics, or memory clinics should be considered when necessary.

[0016] The mobile display module is used to showcase system functions in the form of a web-based calculator, WeChat mini-program, or other mobile devices. The interface includes a homepage description, a rapid screening version, a standard assessment version, a variable input page, a scale auto-scoring page, a results output page, a stratified nursing suggestion page, and a disclaimer page. The rapid screening version is used for large-scale initial screening at the grassroots level, while the standard assessment version is used for further assessment of key populations.

[0017] This invention also provides a method for predicting the risk of mild cognitive impairment and assisting in tiered nursing management among rural elderly, comprising the following steps: collecting multidimensional health variables of rural elderly; entering and automatically scoring scales such as GDS-15, IADL, PSQI, and SSRS; performing format verification, missing value indication, variable coding, and model adaptation processing on the collected variables and scoring results; inputting the processed variables into a pre-trained machine learning model to obtain the predicted probability of mild cognitive impairment risk; classifying individuals into low-risk, medium-risk, or high-risk categories based on preset or locally calibrated risk thresholds; outputting key risk factor prompts based on the importance of model features or interpretation rules; generating corresponding tiered nursing management suggestions and follow-up frequency suggestions according to the risk level; displaying the risk probability, risk level, risk interpretation, and nursing management suggestions through a web calculator or WeChat mini-program, and indicating that the results are only for screening assistance and risk communication. Beneficial effects

[0018] This invention integrates demographic, functional status, psychological sleep, chronic disease, and social behavior factors of rural elderly people, avoiding reliance on a single cognitive scale or human experience judgment.

[0019] This invention transforms the manual calculation of scales such as GDS-15, IADL, PSQI, and SSRS into an itemized collection and automatic scoring process, which can reduce manual scoring errors at the grassroots level and improve the continuity, standardization, and scalability of the screening process.

[0020] This invention outputs individualized risk probabilities through machine learning models, improving the efficiency of risk identification in grassroots settings and facilitating screening assistance and dynamic follow-up for key individuals by nursing staff.

[0021] This invention stratifies risks into low, medium, and high risks based on predicted probabilities and connects the model results with stratified care recommendations, making the risk assessment results more operable.

[0022] This invention provides two application paths: a rapid screening version and a standard assessment version, which can balance the convenience of large-scale initial screening at the grassroots level with the accuracy of further assessment for key populations.

[0023] This invention enables mobile display via a web-based calculator or WeChat mini-program, providing an intuitive presentation of the process from variable input, automatic scale scoring, risk probability output, risk level classification, and nursing recommendation generation, thereby improving the understanding of results among primary care nurses, the elderly, and their families.

[0024] This invention is clearly positioned as an auxiliary tool for risk screening and stratified nursing management, and does not replace clinical diagnosis. It helps to reduce ethical and clinical risks in practical applications. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the overall system structure of the present invention, illustrating the relationship between the data acquisition module, data preprocessing module, nursing assessment and automatic scale scoring module, risk prediction module, risk stratification module, interpretation output module, nursing suggestion generation module, mobile terminal display module, and optional data security and log management module.

[0026] Figure 2 This is a flowchart of the risk prediction and stratified nursing management method of the present invention, showing the complete process from itemized variable collection, preprocessing, automatic scale scoring, model prediction, risk stratification, interpretation output to nursing suggestion generation and mobile terminal display.

[0027] Figure 3 This is a schematic diagram of the functional modules of the web calculator and mini-program interface of the present invention, showing the homepage description, quick screening version, standard assessment version, automatic scale scoring, model prediction results, input information summary, and stratified nursing suggestion output interface.

[0028] Figure 4 This is a risk stratification nursing management logic diagram of the present invention, showing the management focus and reassessment frequency for low-risk, medium-risk and high-risk groups. Detailed Implementation

[0029] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. Without departing from the spirit of the present invention, those skilled in the art can make adaptive adjustments to the variable items, model types, threshold settings, interface forms, scale scoring rules, and nursing recommendations.

[0030] In one embodiment, the system is deployed as a WeChat mini-program. After entering the mini-program, users first read the instructions, clarifying that this tool is for scientific research demonstration and primary care screening assistance, and should not be used as a basis for clinical diagnosis. Subsequently, users can choose between a rapid screening version or a standard assessment version. The rapid screening version collects simple variables such as age, years of education, body mass index, chronic disease status, sensory function, regular exercise, social participation, and cultural, recreational, or intellectually stimulating activities to determine whether further evaluation is recommended. The standard assessment version further embeds items from scales such as GDS-15, IADL, PSQI, and SSRS on top of the rapid screening. Nursing staff select or fill in these items based on interview results, and the system automatically completes the scoring and calls the risk prediction module.

[0031] In another embodiment, the system is deployed as a web-based calculator. The web interface collects variables through sliders, dropdown menus, single-choice, multiple-choice, and tables, and displays automatic scale scoring, predicted probabilities, risk levels, input information summaries, and the basis for tiered nursing management on the same interface. The web interface and the mini-program interface can share the same model parameters or obtain prediction results through a backend model interface.

[0032] In a preferred embodiment, the risk prediction module uses the XGBoost model as the primary risk prediction model. Model input variables include age, years of education, body mass index, total score of instrumental activities of daily living, total sleep quality score, total score of the Geriatric Depression Scale, total score of the Social Support Rating Scale, number of hypertension, diabetes, coronary heart disease or other cardiovascular diseases, number of chronic diseases, hearing problems, vision problems, regular exercise, frequency of social participation, and frequency of cultural, recreational, or intellectually stimulating activities.

[0033] In the study sample, the XGBoost model achieved a mean AUC of 0.938, a mean Brier score of 0.100, a calibration intercept of -0.419, and a calibration slope of 0.781 based on the region-based leave-one-out internal-external validation combined with out-of-the-box prediction results. The Bootstrap resampling population AUC was 0.928, with a 95% confidence interval of 0.905 to 0.948, and the Brier score was 0.100, with a 95% confidence interval of 0.082 to 0.119. These results demonstrate that the model has good discriminative ability and overall prediction error performance. Recalibration can still be performed using local samples during formal deployment.

[0034] Risk stratification thresholds can be determined based on decision curve analysis and candidate threshold classification performance. In one implementation, a predicted probability less than 0.20 is defined as low risk, 0.20 to 0.39 as medium risk, and a predicted probability greater than or equal to 0.40 as high risk. A threshold of 0.20 can be used as an expanded screening threshold, and a threshold of 0.40 can be used as a high-risk identification threshold.

[0035] Example 1: A 68-year-old rural elderly person with 6 years of education, a body mass index of 24 kg / m², an instrumental activities of daily living score of 12, a sleep quality score of 7, a Geriatric Depression Scale score of 5, and a social support assessment score of 35. They have hypertension but no diabetes or coronary heart disease, no hearing or vision problems, exercise regularly, and participate in social and cultural activities only occasionally. The system can output a low-risk or near-medium-risk prediction and recommend routine health education, chronic disease management, and regular follow-up assessments.

[0036] Example 2: A 72-year-old rural elderly person with 3 years of education, a body mass index of 22 kg / m², an instrumental activities of daily living score of 16, a sleep quality score of 9, a Geriatric Depression Scale score of 8, and a social support assessment score of 33. They have hypertension, hearing and vision problems, lack regular exercise, and have limited social participation and cultural and recreational activities. The system can output a high-risk warning and recommend further cognitive screening or specialist evaluation, and collaborate with family members to implement continuous care management.

[0037] The above examples are only for illustrating the system input and output formats. The actual predicted probabilities are subject to the results of the formally deployed model and local recalibration.

[0038] This invention is intended to assist in the risk screening and tiered nursing management of mild cognitive impairment among rural elderly individuals, and is not intended for the sole diagnosis of mild cognitive impairment or to replace physician judgment. For high-risk individuals, a comprehensive confirmation process still requires standardized cognitive assessment, medical history taking, functional assessment, and necessary specialist judgment.

[0039] Before the system is officially rolled out, local recalibration, scale authorization verification, automatic scoring rule validation, user usability testing, prospective validation, and real-world application evaluation should be conducted. If personal health information storage and sharing are involved, the principles of data minimization, informed consent, anonymization, access control, and secure storage should be followed.

Claims

1. A risk prediction and tiered nursing management assistance system for mild cognitive impairment in rural elderly people, characterized in that, include: The application path selection module provides two usage paths: a quick screening version and a standard evaluation version. The data acquisition module is used to collect information on multidimensional health variables and standardized scale items of rural elderly people; The data preprocessing module is used to perform format verification, missing value indication, variable coding, and model adaptation processing on the multidimensional health variables and scale scoring results; the nursing assessment and automatic scale scoring module is used to automatically generate the total scale score or dimension score based on the selection or completion results of standardized scale items; the risk prediction module is used to input the processed variables into a pre-trained machine learning model and output the risk prediction probability of mild cognitive impairment; the risk stratification module is used to classify individuals into low-risk, medium-risk, or high-risk categories based on the risk prediction probability. The output module is used to output key risk factors related to the risk prediction results. The nursing suggestion generation module is used to generate corresponding stratified nursing management suggestions based on risk levels. It also includes a mobile display module, which is used to display variable input, automatic scale scoring, risk prediction probability, risk level, and stratified nursing management suggestions through a web calculator, WeChat mini-program, or mobile terminal interface.

2. The system according to claim 1, characterized in that, The rapid screening version is used to collect one or more simple variables, including age, years of education, body mass index, chronic disease status, sensory function, regular exercise, social participation, frequency of cultural, recreational or intellectual activities, and subjective memory decline, to determine whether to recommend moving to the standard assessment version. The standard assessment version is used to further collect information on one or more scale items from GDS-15, IADL, PSQI and SSRS based on the rapid screening, and the system automatically completes the scoring.

3. The system according to claim 1, characterized in that, The nursing assessment and scale automatic scoring module includes one or more of the following: GDS-15 automatic scoring unit, IADL automatic scoring unit, PSQI automatic scoring unit, and SSRS automatic scoring unit. Specifically, the GDS-15 automatic scoring unit is used to convert and summarize preset reverse scoring items; the IADL automatic scoring unit is used to calculate the total score based on instrumental activities of daily living (ADL) items; the PSQI automatic scoring unit is used to convert sleep-related items into seven component scores and calculate the total score; and the SSRS automatic scoring unit is used to calculate objective support, subjective support, support utilization, and the total score based on social support items.

4. The system according to claim 1, characterized in that, The multidimensional health variables include basic information, scales and functional status, chronic diseases and health status, and behavioral and social factors. The basic information includes age, years of education, body mass index, and average monthly household income per capita. The scales and functional status include the total score of instrumental activities of daily living, the total score of sleep quality, the total score of the Geriatric Depression Scale, and the total score of the Social Support Rating Scale. The chronic diseases and health status include hypertension, diabetes, coronary heart disease or other cardiovascular diseases, the number of chronic diseases, hearing problems, and vision problems. The behavioral and social factors include regular exercise, frequency of social participation, and frequency of cultural, recreational, or intellectually stimulating activities.

5. The system according to claim 1, characterized in that, The data preprocessing module includes a numerical range verification unit, a missing variable prompting unit, a binary variable encoding unit, an ordered categorical variable encoding unit, and a continuous variable standardization or model adaptation unit.

6. The system according to claim 1, characterized in that, The pre-trained machine learning model is an XGBoost model, a random forest model, an elastic network model, a logistic regression model, or other classification models that can output probability values; among them, the XGBoost model is used as the main risk prediction model, and the logistic regression model or noctilinear plot is used as an auxiliary explanatory and display model.

7. The system according to claim 1, characterized in that, The risk stratification module includes a first threshold and a second threshold. The first threshold is used to distinguish between low risk and medium risk, and the second threshold is used to distinguish between medium risk and high risk. When the risk prediction probability is lower than the first threshold, it is classified as low risk. When the risk prediction probability is greater than or equal to the first threshold and lower than the second threshold, it is classified as medium risk. When the risk prediction probability is greater than or equal to the second threshold, it is classified as high risk.

8. The system according to claim 7, characterized in that, The first threshold is 0.20, and the second threshold is 0.40; or, the first and second thresholds are recalibrated based on local samples, model calibration results, or decision curve analysis results.

9. The system according to claim 1, characterized in that, The explanation output module outputs factors related to increased risk based on the importance of model features, SHAP value, or preset explanation rules. These factors include one or more of the following: decreased instrumental daily living activities, shorter years of education, abnormal body mass index, poor sleep quality, significant depressive symptoms, heavy burden of chronic diseases, sensory function problems, insufficient social support, or limited social participation.

10. The system according to claim 1, characterized in that, The nursing recommendation generation module is configured to: generate recommendations for cognitive health education, lifestyle guidance, routine chronic disease management, and 6-12 month reassessment for low-risk individuals; generate recommendations for key follow-up, sleep management, mood screening, chronic disease management, functional status monitoring, social support promotion, and 3-6 month reassessment for medium-risk individuals; and generate recommendations for further cognitive screening or specialist assessment, individualized nursing plans, continuous comprehensive nursing, necessary referrals, and 1-3 month reassessment for high-risk individuals.

11. The system according to claim 1, characterized in that, The mobile terminal display module includes a homepage description unit, a quick screening unit, a standard assessment unit, a scale automatic scoring display unit, a result output unit, a risk stratification basis display unit, a nursing suggestion display unit, and a disclaimer unit. The disclaimer unit is used to indicate that the system results are only used as an aid to risk screening and stratified nursing management and do not replace clinical diagnosis.

12. The system according to claim 1, characterized in that, It also includes a data security and log management module, which is used to record model version information, user authorization status, anonymized input records and system operation logs, and to implement data minimization, access control and secure storage.

13. A method for risk prediction and tiered nursing management of mild cognitive impairment in rural elderly, characterized in that, The process includes the following steps: providing two application paths, a rapid screening version and a standard assessment version; collecting information on multidimensional health variables and standardized scale items among rural elderly people; and automatically scoring one or more scales, such as GDS-15, IADL, PSQI, and SSRS. The multidimensional health variables and scale scores are processed for format validation, missing value indication, variable encoding, and model adaptation. The processed variables are then input into a pre-trained machine learning model to obtain the probability of predicting the risk of mild cognitive impairment. Individuals are classified as low-risk, medium-risk, or high-risk based on preset risk thresholds or locally calibrated risk thresholds; key risk factors are indicated based on the importance of model features or interpretation rules. Based on the risk level, generate corresponding tiered nursing management recommendations and reassessment frequency recommendations; It also displays risk prediction probability, risk level, risk factor tips, and tiered nursing management suggestions through web calculators, WeChat mini programs, or mobile interfaces.

14. The method according to claim 13, characterized in that, The automatic scoring includes: converting and summarizing the GDS-15 preset reverse scoring items; summing the IADL items to obtain the total score for instrumental daily living activities; converting the PSQI items into seven component scores and calculating the total score from 0 to 21; and calculating the objective support, subjective support, support utilization, and total score for the SSRS items.

15. The method according to claim 13, characterized in that, The multidimensional health variables include one or more of the following: age, years of education, body mass index, average monthly household income per capita, total score of instrumental activities of daily living, total score of sleep quality, total score of the Geriatric Depression Scale, total score of the Social Support Rating Scale, hypertension, diabetes, coronary heart disease or other cardiovascular diseases, number of chronic diseases, hearing problems, vision problems, regular exercise, frequency of social participation, and frequency of cultural, recreational or intellectual activities.

16. The method according to claim 13, characterized in that, The risk thresholds include a first threshold and a second threshold, which can be determined based on local samples, model calibration results, or decision curve analysis results. In one embodiment, the first threshold is 0.20 and the second threshold is 0.

40. When the risk prediction probability is less than 0.20, it is classified as low risk; when the risk prediction probability is between 0.20 and 0.39, it is classified as medium risk; and when the risk prediction probability is greater than or equal to 0.40, it is classified as high risk.

17. The method according to claim 13, characterized in that, The generated tiered nursing management recommendations include: automatically matching at least one of the following based on risk level: health education, follow-up frequency, sleep management, mood screening, chronic disease management, functional maintenance training, family support, and referral recommendations.

18. The method according to claim 13, characterized in that, When displaying the predicted risk probability, risk level, risk factor indications, and stratified nursing management recommendations, the system also displays a boundary indication of the results’ use. This boundary indication states that the system results are only for primary care screening assistance, risk communication, and stratified nursing management reference, and should not be used as a sole basis for diagnosis.

19. An electronic device comprising a processor and a memory, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the method according to any one of claims 13 to 18.

20. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 13 to 18.

Images