Clinical model for predicting cognitive impairment risk of old people in rural areas in China and design method

Abstract

The invention discloses a cognitive impairment risk prediction clinical model for Chinese rural old people and a design method, and belongs to the crossing field of senile pathology and medical artificial intelligence. The input module is used for receiving individual data including age, education level, drinking behavior, systolic pressure, grip strength and depression state; the processing module is used for screening variables based on LASSO regression and calculating the total score T, the prediction score LogitP and the probability P of the cognitive impairment of the individual through multi-factor Logistic regression; and the output module is used for outputting the calculation result of the processing module for clinical use. The model is based on a CHARLS database, integrates demographic characteristics, lifestyle, chronic disease history and body function indexes, AUC in a training set and AUC in a verification set are 0.849 and 0.852 respectively, and the model has good discrimination ability, calibration degree and clinical practicability and can provide a convenient and efficient cognitive impairment risk screening tool for primary medical institutions.

Description

Technical Field

[0001] The present invention belongs to the intersection of geriatrics and medical artificial intelligence, and specifically relates to a clinical model and design method for predicting the risk of cognitive impairment in the elderly population in rural China, which is used for early screening and decision-making support for cognitive dysfunction in the elderly population in rural communities in China. Background Art

[0002] Cognitive impairment has become a major public health challenge threatening the health and quality of life of the elderly. As a syndrome ranging from mild cognitive impairment (MCI) to various types of dementia, the scale of cognitive impairment cannot be ignored. A 2020 national epidemiological study showed that approximately 38.77 million people aged 60 and above suffer from MCI, and another 15.07 million suffer from dementia, with the overall prevalence on the rise. However, most existing studies have focused on urban areas, and the research samples are mostly from groups with higher education levels or better socioeconomic conditions. Insufficient attention has been paid to the epidemiological characteristics and risk factors of cognitive impairment in rural elderly people.

[0003] It is worth noting that the prevalence of cognitive impairment among the elderly in rural areas is generally higher than that in cities. Some studies have pointed out that their risk of disease is 1.3 to 1.5 times higher than that of the elderly in urban areas, suggesting that this group faces more serious cognitive health problems and urgently needs to strengthen early screening and intervention measures to slow the progression of the disease and reduce the burden of social care.

[0004] Early identification of individuals at high risk for cognitive impairment is crucial for delaying cognitive decline, optimizing health resource allocation, and developing individualized intervention strategies. Currently, researchers at home and abroad have developed a variety of cognitive impairment prediction tools. However, for elderly people in rural communities, there is still a lack of predictive models based on large-scale Chinese sample data, rigorous methodology, and good clinical operability.

[0005] The China Health and Retirement Longitudinal Study (CHARLS), a nationwide longitudinal cohort study, systematically collects multidimensional information on the health, socioeconomic, and behavioral characteristics of the elderly population, providing a solid data foundation for the development of cognitive impairment risk assessment tools. This study leverages the 2011 CHARLS database, integrating demographic characteristics, lifestyle, chronic disease history, and physical function variables. LASSO regression is used for feature selection, and multivariate logistic regression is combined to develop a nomogram model. The nomogram's discriminatory ability, calibration, and potential clinical application are systematically evaluated. The aim is to provide a practical, quantitative aid for the early identification and precise management of cognitive impairment in elderly people in rural areas. Summary of the Invention

[0006] The purpose of the present invention is to propose a clinical model and design method for predicting the risk of cognitive impairment in the elderly population in rural China. The model is constructed based on the large-sample, nationally representative CHARLS database, has good predictive performance and clinical practicality, and can be used for early screening and risk management of cognitive dysfunction in the elderly population in grassroots communities.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is:

[0008] A clinical model for predicting the risk of cognitive impairment in the elderly population in rural China, including:

[0009] An input module for receiving individual data, including age, education level, drinking behavior, systolic blood pressure, grip strength, and depression status;

[0010] The processing module screens variables based on LASSO regression and calculates the total score T, the predicted score LogitP, and the probability P of an individual developing cognitive impairment through multivariate logistic regression:

[0011] T = A drinking behavior + B depressive status + C education level + D systolic blood pressure + E age + F handgrip strength (1)

[0012] Logit(P) = -4.69×10 -7 * T 3 + 0.000314672 * T 2 - 0.062407352 * T +3.955679447 (2)

[0013] (3)

[0014] In formula (1),

[0015] When drinking occurs, A is 0; when drinking does not occur, A is 12.21;

[0016] When there is a depressive state, B is 28.93; when there is no depressive state, B is 0;

[0017] When the education level is below primary school, C is 105; when the education level is primary school, C is 100; when the education level is junior high school, C is 61.21; when the education level is high school or above, C is 17.27;

[0018] When systolic blood pressure (SBP) is x (mmHg), D = 0.3169 *x - 25.3549;

[0019] When age is y = 60 to 105 (years), E = 2.0103*y - 120.6151;

[0020] When the grip force is z (kg), F = -1.1512*z + 69.0699;

[0021] Output module, outputs the calculation results of the processing module for clinical use.

[0022] Furthermore, the present invention also proposes a design method for the predictive clinical model, which comprises the following steps:

[0023] ① LASSO regression was used with 10-fold cross-validation to determine the optimal penalty parameter λ. Thirteen variables with non-zero regression coefficients were screened, namely risk factors: age, education level, sleep duration, social activities, ADLs, IADLs, grip strength, alcohol consumption, systolic blood pressure, depression, disability, self-rated health, and tooth loss.

[0024] ②Multivariate logistic regression analysis of the above risk factors showed that age, education level, systolic blood pressure, alcohol consumption, grip strength, and depression were independent risk factors for cognitive impairment in the rural elderly population;

[0025] ③Construct a clinical prediction model for the risk of cognitive impairment in rural elderly people composed of the above independent risk factors.

[0026] After obtaining the clinical prediction model, the area under the receiver operating characteristic (ROC) curve (AUC), calibration curve, and decision curve were used to verify the model's discriminability, goodness of fit, and clinical application value, respectively.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. This invention establishes a clinical prediction model for cognitive impairment in the elderly population in rural China. First, based on the 2011 CHARLS database, 2,228 elderly people aged 60 and above living in rural communities were included. LASSO regression was used to select 38 candidate variables for features. Subsequently, multivariate logistic regression analysis was performed to identify six independent risk factors for cognitive impairment: age, education level, drinking behavior, systolic blood pressure, handgrip strength, and depression. A nomogram prediction model was then established based on these factors. Finally, the model's performance was systematically validated in the modeled population using the area under the receiver operating characteristic (ROC) curve (AUC), calibration curve, and decision curve analysis (DCA). The results demonstrated that the model achieved AUCs of 0.849 and 0.852 in the training and validation sets, respectively, indicating good discriminatory power. The calibration curve demonstrated high consistency between the model's predicted values ​​and the actual incidence rates, and the DCA results further demonstrated the model's high clinical value.

[0029] 2. The prediction model established by the present invention has good discrimination, good fit and clinical practicality, simple structure and convenient variable acquisition. It can be used for early identification and individualized intervention of the risk of cognitive impairment in rural elderly people in grassroots community environments, and has high accuracy and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 The LASSO regression cross-validation and path coefficient plots show the optimal penalty parameter λ selected based on 10-fold cross-validation. The left dashed line is the λ corresponding to the minimum error, and the right dashed line is λ1se.

[0031] Figure 2 The LASSO path diagram shows the trend of each variable coefficient changing with λ, and the dotted line is the optimal λ value selected.

[0032] Figure 3 This is a nomogram for the prediction model.

[0033] Figure 4 This is the prediction performance evaluation graph (ROC curve) of the model in the training set.

[0034] Figure 5 The prediction performance evaluation graph (calibration curve) of the model in the training set.

[0035] Figure 6 This is the prediction performance evaluation graph of the model in the training set (DCA curve).

[0036] Figure 7 This is the prediction performance evaluation graph of the model in the validation set (ROC curve).

[0037] Figure 8 The prediction performance evaluation graph of the model in the validation set (calibration curve).

[0038] Figure 9 This is the prediction performance evaluation graph of the model in the validation set (DCA curve). DETAILED DESCRIPTION

[0039] The present invention is further described below in conjunction with embodiments and drawings.

[0040] 1 Materials and Methods

[0041] 1.1 Research subjects

[0042] This study was based on the 2011 CHARLS baseline data. The survey protocol was approved by the Peking University Biomedical Ethics Committee (IRB00001052-11015), and all participants provided written informed consent. The 2011 baseline survey included 17,708 participants, who were excluded based on the following criteria: ① age under 60 years; ② urban residence; and ③ missing cognitive function score data. A total of 2,228 eligible subjects were included in the study.

[0043] 1.2 Research indicators and data collection

[0044] 1.2.1 Cognitive function assessment

[0045] This study drew on methods from the Health and Retirement Study (HRS) in the United States to assess cognitive function in the CHARLS study population. Cognitive function assessment encompasses two dimensions: mental state and episodic memory. Mental state was assessed using digit recognition, calculation, and figure drawing tasks, with a total score of 11. Episodic memory was tested using a word recall task, with a total score of 20. The sum of these two scores constituted a total cognitive function score of 31. Based on previous research criteria, cognitive impairment is defined as a total cognitive function score below 11.

[0046] 1.2.2 Research variables

[0047] The variables included in this study mainly include demographic and sociological characteristics (gender, age, marital status, education level, life satisfaction), lifestyle and behavioral habits (smoking, drinking, sleep duration, social activities, access to tap water), history of chronic diseases (hypertension, diabetes, cancer, lung disease, heart disease, stroke, arthritis, dyslipidemia, liver disease, kidney disease, stomach disease, asthma, hip fracture, visual impairment, hearing impairment, depression, pain), physical function and subjective health status (self-rated health, ADL, IADL, waist circumference, BMI, grip strength, systolic blood pressure, diastolic blood pressure, disability, history of falls, tooth loss).

[0048] 1.3 Statistical methods

[0049] All statistical analyses in this study were performed in R software (version 4.4.1). Continuous variables were expressed as medians with interquartile ranges (IQRs), and between-group comparisons were performed using the Mann–Whitney U test or the Kruskal–Wallis H test. Categorical variables were expressed as frequencies and percentages, and between-group differences were tested using the chi-square test (χ² test) or Fisher's exact test. To identify potential influencing factors associated with cognitive impairment in the elderly, LASSO regression was first used for variable selection. Subsequently, multivariate logistic regression analysis was performed to identify independent risk factors. A nomogram was constructed based on the regression results to visualize the model. The receiver operating characteristic (ROC) curve was used to analyze the discriminative ability of the model, and the area under the curve (AUC) was used to measure predictive accuracy. Calibration curves were used to assess the consistency between the model's predicted values ​​and the observed values. Decision curve analysis (DCA) was also used to evaluate the clinical utility of the model at different risk thresholds. All statistical tests were two-sided, and differences were considered statistically significant at P < 0.05.

[0050] 2 Results

[0051] 2.1 Basic Situation

[0052] This study included 1560 subjects in the training set and 668 subjects in the validation set, with the prevalence of cognitive impairment being 17.2% and 17.4%, respectively. In the training set, several baseline characteristics were significantly different between the cognitive impairment and non-cognitive impairment groups (P < 0.05), including gender, age, marital status, education level, waist circumference, body mass index (BMI), systolic blood pressure, diastolic blood pressure, grip strength, sleep duration, social activity participation, self-rated health status, activities of daily living (ADL), instrumental activities of daily living (IADL), tap water use, alcohol consumption, hypertension, dyslipidemia, gastric disease, depressive symptoms, chronic pain, physical disability, history of falls, vision and hearing impairment, and tooth loss. See Table 1 for details.

[0053] Table 1 Baseline characteristics of participants in the training set

[0054] variable <![CDATA[No cognitive impairment (n = 1,331) 1 > <![CDATA[Cognitive impairment (n = 229) 1 > <![CDATA[P value 2 > gender female 601(45.2) 147(64.2) <0.001 male 730(54.8) 82(35.8) age 66(62,71) 70(64,76) <0.001 marriage unmarried 205(15.4) 66(28.8) <0.001 Married or other 1,126(84.6) 163(71.2) Education level Primary school and below 405(30.4) 182(79.5) <0.001 primary school 375(28.2) 38(16.6) junior high school 304(22.8) 6(2.6) High school and above 247(18.6) 3(1.3) Life satisfaction Low 119(8.9) 38(16.6) 0.002 middle 911(68.4) 141(61.6) high 301(22.6) 50(21.8) Smoking no 971(73.0) 172(75.1) 0.496 yes 360(27.0) 57(24.9) drinking no 897(67.4) 187(81.7) <0.001 yes 434(32.6) 42(18.3) Sleep time <6h 368(27.6) 89(38.9) <0.001 6-8h 887(66.6) 117(51.1) >8h 76(5.7) 23(10.0) social activities none 583(43.8) 135(59.0) <0.001 have 748(56.2) 94(41.0) There is running water 234(17.6) 69(30.1) <0.001 1,097(82.4) 160(69.9) hypertension no 559(42.0) 73(31.9) 0.004 yes 772(58.0) 156(68.1) diabetes no 1,095(82.3) 192(83.8) 0.563 yes 236(17.7) 37(16.2) cancer no 1,313(98.6) 227(99.1) 0.756 yes 18(1.4) 2(0.9) pulmonary disease no 1,166(87.6) 197(86.0) 0.507 yes 165(12.4) 32(14.0) heart disease no 1,035(77.8) 183(79.9) 0.467 yes 296(22.2) 46(20.1) stroke no 1,277(95.9) 220(96.1) 0.928 yes 54(4.1) 9(3.9) arthritis no 898(67.5) 146(63.8) 0.270 yes 433(32.5) 83(36.2) Dyslipidemia no 1,083(81.4) 203(88.6) 0.007 yes 248(18.6) 26(11.4) Liver disease no 1,271(95.5) 222(96.9) 0.317 yes 60(4.5) 7(3.1) Kidney disease no 1,246(93.6) 215(93.9) 0.876 yes 85(6.4) 14(6.1) stomach problems no 1,066(80.1) 170(74.2) 0.044 yes 265(19.9) 59(25.8) Asthma no 1,260(94.7) 218(95.2) 0.740 yes 71(5.3) 11(4.8) Hip fracture no 1,315(98.8) 223(97.4) 0.120 yes 16(1.2) 6(2.6) visual impairment no 324(24.3) 35(15.3) 0.003 yes 1,007(75.7) 194(84.7) Hearing impairment no 606(45.5) 83(36.2) 0.009 yes 725(54.5) 146(63.8) depression no 975(73.3) 104(45.4) <0.001 yes 356(26.7) 125(54.6) pain no 1,033(77.6) 148(64.6) <0.001 yes 298(22.4) 81(35.4) Self-rated health Difference 295(22.2) 74(32.3) 0.001 generally 720(54.1) 117(51.1) good 316(23.7) 38(16.6) ADL 0.00(0.00,0.00) 0.00(0.00,1.00) <0.001 IADL 0.00(0.00,0.00) 0.00(0.00,2.00) <0.001 waistline 87(83,93) 87(81,92) 0.003 BMI 24.6(22.3,25.9) 23.4(20.8,24.8) <0.001 grip strength 29(24,33) 23(19,27) <0.001 Systolic blood pressure 135(124,144) 140(126,153) <0.001 Diastolic blood pressure 76 (70, 82) 77 (71, 84) 0.007 Disability 1,136(85.3) 160(69.9) <0.001 195(14.7) 69(30.1) History of falls 1,133(85.1) 182(79.5) 0.030 198(14.9) 47(20.5) Tooth loss 1,199(90.1) 177(77.3) <0.001 132(9.9) 52(22.7)

[0055] 1 Data are expressed as n (%) or median (Q1, Q3); 2 Comparisons between groups were performed using the Pearson chi-square test, Fisher's exact test, or the Wilcoxon rank-sum test.

[0056] 2.2 Model Construction

[0057] In this study, the LASSO regression model was applied to the training set to reduce the dimension of candidate variables and the optimal penalty coefficient λ was determined through 10-fold cross-validation. The results showed that when λ = 0.0168, the model error was within one standard deviation of its minimum value, so this value was selected as the final parameter. LASSO analysis screened out 13 variables with non-zero coefficients, including age, education level, sleep time, social activities, ADL, IADL, grip strength, alcohol consumption, systolic blood pressure, depression, disability, self-rated health, and tooth loss. Figure 1 、 2 .

[0058] Based on this data, a multivariate logistic regression analysis was performed. The results showed that in the training set (n = 1,560), the following factors were independent influencing factors for cognitive impairment: age, education level, alcohol consumption, systolic blood pressure, handgrip strength, and depression (see Table 2 for details).

[0059] Table 2 Independent risk factors

[0060] project Model coefficient (B) Standard error (SE) OR 95% CI p-value (p) age 0.059 0.013 1.07 1.04, 1.09 <0.001 Education level (referring to primary school and below) (primary school) -1.250 0.207 0.28 0.19, 0.42 <0.001 (junior high school) -2.763 0.440 0.07 0.03, 0.15 <0.001 (High school and above) -3.210 0.597 0.04 0.01, 0.12 <0.001 Drinking (reference) -0.448 0.209 0.67 0.45, 1.00 0.050 Systolic blood pressure 0.011 0.004 1.01 1.00, 1.02 0.007 grip strength -0.032 0.011 0.96 0.94, 0.98 <0.001 Depression (reference no) 0.725 0.180 2.59 1.87, 3.58 <0.001

[0061] Based on LASSO regression, variables were selected and multivariate logistic regression was used to calculate the total score T, the predicted score LogitP, and the probability P of individual cognitive impairment, and a nomogram model was constructed (see Figure 3 ), used to predict the individual risk of cognitive impairment in the elderly population.

[0062] T = A drinking behavior + B depressive status + C education level + D systolic blood pressure + E age + F handgrip strength (1)

[0063] Logit(P) = -4.69×10 -7 * T 3 + 0.000314672 * T 2 - 0.062407352 * T +3.955679447 (2)

[0064] (3)

[0065] In formula (1),

[0066] When drinking occurs, A is 0; when drinking does not occur, A is 12.21;

[0067] When there is a depressive state, B is 28.93; when there is no depressive state, B is 0;

[0068] When the education level is below primary school, C is 105; when the education level is primary school, C is 100; when the education level is junior high school, C is 61.21; when the education level is high school or above, C is 17.27;

[0069] When systolic blood pressure (SBP) is x (mmHg), D = 0.3169 *x - 25.3549;

[0070] When age is y = 60 to 105 (years), E = 2.0103*y - 120.6151;

[0071] When the grip force is z (kg), F = -1.1512*z + 69.0699;

[0072] 2.3 Internal and external validation of the nomogram prediction model

[0073] Based on the 6 independent influencing factors screened out, a cognitive impairment risk prediction model suitable for the elderly population was constructed. The model performance was evaluated using the training set and validation set. The results showed that its AUC was 0.849 (95% CI: 0.825-0.873) and 0.852 (95% CI: 0.817-0.888), respectively, indicating that the model has good discrimination ability. The calibration curve results showed that the predicted probability of the training set and the validation set was highly consistent with the actual incidence rate. The fitting curve was close to the ideal diagonal line, and the slope was approximately 1, indicating that the model has good calibration performance. In addition, DCA showed that within a wider risk threshold range, the nomogram prediction model can achieve higher net clinical benefits and has strong clinical practicality and promotion and application value. Figure 4-9 .

[0074] 2.4 Establishment of a clinical model for predicting cognitive impairment risk in the elderly population in rural China

[0075] The input module is used to receive individual data, including age, education level, drinking behavior, systolic blood pressure, grip strength, and depression status.

[0076] In the processing module, variables are screened based on LASSO regression and multivariate logistic regression is used to calculate the total score T, the predicted score LogitP, and the probability P of an individual developing cognitive impairment. The calculation formula is shown in step 2.2.

[0077] Output module, outputs the calculation results of the processing module for clinical use.

Claims

1. A clinical model for predicting the risk of cognitive impairment in the elderly population in rural China, characterized by: include: An input module for receiving individual data, including age, education level, drinking behavior, systolic blood pressure, grip strength, and depression status; The processing module screens variables based on LASSO regression and calculates the total score T, the predicted score LogitP, and the probability P of an individual developing cognitive impairment through multivariate logistic regression: T = A drinking behavior + B depressive status + C education level + D systolic blood pressure + E age + F handgrip strength (1) Logit(P) = - 4.69×10 -7 * T 3 + 0.000314672 * T 2 - 0.062407352 * T +3.955679447 (2) （3） In formula (1), When drinking occurs, A is 0; when drinking does not occur, A is 12.21; When there is a depressive state, B is 28.93; when there is no depressive state, B is 0; When the education level is below primary school, C is 105; when the education level is primary school, C is 100; when the education level is junior high school, C is 61.21; when the education level is high school or above, C is 17.27; When systolic blood pressure (SBP) is x (mmHg), D = 0.3169 *x - 25.3549; When age is y = 60 to 105 (years), E = 2.0103*y - 120.6151; When the grip force is z (kg), F = -1.1512*z + 69.0699; Output module, outputs the calculation results of the processing module for clinical use.

2. The method for designing a clinical model for predicting the risk of cognitive impairment in the elderly population in rural China as claimed in claim 1, characterized in that: Here are the steps: ① LASSO regression was used with 10-fold cross-validation to determine the optimal penalty parameter λ. Thirteen variables with non-zero regression coefficients were screened, namely risk factors: age, education level, sleep duration, social activities, ADLs, IADLs, grip strength, alcohol consumption, systolic blood pressure, depression, disability, self-rated health, and tooth loss. ②Multivariate logistic regression analysis of the above risk factors showed that age, education level, systolic blood pressure, alcohol consumption, grip strength, and depression were independent risk factors for cognitive impairment in the rural elderly population; ③Construct a clinical prediction model for the risk of cognitive impairment in rural elderly people composed of the above independent risk factors.

3. The design method according to claim 2, wherein: After obtaining the clinical prediction model, the area under the receiver operating characteristic (ROC) curve (AUC), calibration curve, and decision curve were used to verify the model's discriminability, goodness of fit, and clinical application value, respectively.

Images