Construction method and application of cardiovascular metabolism co-disease elderly patient disability risk prediction model

Abstract

The invention relates to the related technical field of prediction models, in particular to a construction method and application of a cardiovascular metabolism co-disease elderly patient disability risk prediction model based on a dynamic column diagram. The method comprises the following steps: screening data of cardiovascular metabolism co-disease elderly patients by using a CHARLS database, including disability risk factors and outcome indexes; screening through Logistic regression analysis and Lasso regression analysis to obtain final prediction variables, and constructing a visual dynamic column graph prediction model by using a shiyPredict packet based on the prediction variables; performing internal verification by using a bootstrap party, and performing external verification by using data of different years in the CHARLS database; and respectively evaluating the distinguishing capability, the calibration capability and the clinical effectiveness of the model through an ROC curve, a calibration curve and a DCA curve. The visual prediction model can help clinicians to identify high-disability-risk patients in an early stage and perform intervention in time, so that the disability occurrence rate is reduced, and short-term and long-term prognosis of cardiovascular metabolism co-disease elderly patients is improved.

Description

Technical Field

[0001] The present invention relates to the technology of constructing prediction models, belonging to the field of new generation information technology. Specifically, it relates to a method for constructing and applying a disability risk prediction model for elderly patients with cardiovascular-metabolic multimorbidity based on a dynamic nomogram. Background Art

[0002] Cardiovascular-metabolic multimorbidity (CMM) is defined as an individual having two or more of the following diseases simultaneously: hypertension, heart disease, stroke, diabetes, and dyslipidemia. With the intensification of global population aging, the prevalence of CMM is continuously increasing in the elderly population. Elderly patients with CMM are prone to serious complications due to the interaction of multiple diseases leading to impaired multi-organ function. At the same time, the co-use of multiple drugs brings side effects, and poor treatment compliance affects disease control. These factors together accelerate the decline of physical function and increase the risk of disability. Disability may lead to various adverse events such as falls, hospitalizations, and deaths, thereby reducing the individual's self-care ability and quality of life, and bringing a heavy economic burden to the medical system, family, and society.

[0003] Currently, most of the screening and assessment tools for disability are general tools, and there is still a lack of screening tools that are suitable for clinical medical staff, relatively objective, convenient, and targeted at the disability of elderly patients with CMM. Moreover, most research indicators are single, lacking a prediction model that combines multiple indicators.

[0004] Therefore, there is an urgent need to develop a disability risk prediction model for elderly patients with CMM that is constructed based on clinical routine test items and combines multiple indicators, so as to improve the scientific nature of predicting the disability risk of elderly patients with CMM and reduce the incidence of disability risk in elderly patients with CMM. Summary of the Invention

[0005] In view of the above problems, the purpose of the present invention is to construct a prediction model based on a dynamic nomogram, and predict the disability probability of elderly patients with CMM through the model, so as to more intuitively and conveniently detect high-risk disability patients early and intervene in a timely manner to prevent the occurrence of disability hazards.

[0006] To achieve the above purpose, the technical solution of the present invention is as follows:

[0007] A method for constructing a disability risk prediction model for elderly patients with CMM based on a dynamic nomogram, comprising the following steps:

[0008] S1. Screening data and preprocessing: Screening candidate prediction variables of elderly patients with CMM from the CHARLS database according to inclusion and exclusion criteria and performing preprocessing;

[0009] S2. Screening of predictive variables: Screen the final predictive variables related to the disability risk of elderly CMM patients;

[0010] S3. Construction of a prediction model: Use the predictive variables as independent influencing factors to construct a nomogram and a web-based dynamic nomogram prediction model;

[0011] S4. Internal and external validation and evaluation of the model.

[0012] Furthermore, the candidate predictive variables include 1. Demographic and mental health variables 2. Lifestyle and health behavior variables 3. Laboratory test variables 4. Physical examination variables 5. Clinical-related variables.

[0013] Furthermore, in step S1, the inclusion and exclusion criteria are as follows: Include the fasting CMM population aged ≥ 60 years in the CHARLS database in 2011 and 2015 and containing 47 relevant variables; Exclude those who do not meet the inclusion criteria and those with missing values in the 47 relevant variables; Based on the R software (version 4.1.0), randomly split the data in the CHARLS database in 2011 and 2015 into a training set and an internal validation set at a ratio of 7:3.

[0014] Furthermore, in step S2, the candidate predictive variables are analyzed by univariate Logistic regression, Lasso regression, and multivariate Logistic regression to obtain the final predictive variables and establish a nomogram prediction model.

[0015] Furthermore, in step S3, a web-based dynamic nomogram is developed based on the shinyPredict package of the R software to achieve real-time interactive risk prediction.

[0016] Furthermore, in step S4, the Bootstrap method is used for internal validation, and the data of 3335 elderly CMM patients containing the final predictive variables in the CHARLS database in 2018 and 2020 are used to complete the external validation.

[0017] Furthermore, in step S4, the performance of the model is evaluated through ROC curves, calibration curves, and DCA curves.

[0018] Compared with the prior art, the innovative advantages of this study are as follows:

[0019] 1. The present invention is developed based on the Chinese CHARLS database but has cross - regional applicability. The dynamic nomogram technology supports multi - platform access, providing a standardized risk assessment tool for different medical scenarios. It develops a disability prediction model for the elderly CMM patient group for the first time. The disability risk in this population is significantly higher than that of non - comorbid patients, filling the research gap in this field. In addition, the present invention selects 46 candidate prediction variables, which comprehensively integrate information in multiple aspects such as demographics, lifestyle, and clinical characteristics, thus ensuring the comprehensiveness of the disability risk assessment.

[0020] 2. An online nomogram is constructed using dynamic visualization technology. The model can achieve real - time interactive input of clinical parameters and dynamic output of risk probabilities, improving clinical usability compared with traditional static models. The model can identify high - risk individuals at the initial diagnosis stage, providing new strategies for early identification of disability and precise intervention, thereby reducing the incidence of disability in elderly CMM patients, improving prognosis, and enhancing the quality of life. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The embodiments of the present invention will be described in more detail by combining the accompanying drawings. The above - mentioned and other objects, features, and advantages of the present invention will become more obvious. The drawings are used to provide further understanding of the embodiments of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings, the same reference numerals generally represent the same components or steps.

[0022] FIG. 1 is a schematic flow chart of the construction method and application of the disability risk prediction model for elderly CMM patients of the present invention;

[0023] FIG. 2 is a detailed inclusion - exclusion flow chart of the research data of the present invention;

[0024] FIG. 3 is a process diagram of variable screening by increasing lambda in the Lasso regression model of the present invention;

[0025] FIG. 4 is a distribution diagram of the Lasso coefficients of 16 candidate prediction variables of the present invention;

[0026] FIG. 5 is a demonstration diagram of the application of the disability risk nomogram prediction model for elderly CMM patients of the present invention;

[0027] FIG. 6 is a network - version dynamic nomogram prediction model diagram of the disability risk for elderly CMM patients of the present invention;

[0028] FIG. 7 is a demonstration diagram of the application of the disability risk dynamic nomogram prediction model for elderly CMM patients of the present invention;

[0029] FIG. 8 is a ROC curve diagram of the training set for predicting disability in elderly CMM patients of the present invention;

[0030] Figure 9 is the ROC curve of the internal validation set for predicting disability in elderly CMM patients in the present invention;

[0031] Figure 10 is the ROC curve of the external validation set for predicting disability in elderly CMM patients in the present invention;

[0032] Figure 11 is the calibration curve of the training set for predicting disability in elderly CMM patients in the present invention;

[0033] Figure 12 is the calibration curve of the internal validation set for predicting disability in elderly CMM patients in the present invention;

[0034] Figure 13 is the calibration curve of the external validation set for predicting disability in elderly CMM patients in the present invention;

[0035] Figure 14 is the DCA curve of the training set for predicting disability in elderly CMM patients in the present invention;

[0036] Figure 15 is the DCA curve of the internal validation set for predicting disability in elderly CMM patients in the present invention;

[0037] Figure 16 is the DCA curve of the external validation set for predicting disability in elderly CMM patients in the present invention. Detailed implementation manners

[0038] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0039] Please refer to Tables 1-3 and Figures 1-16 , the present invention is based on a dynamic nomogram prediction model for online dynamically predicting the disability risk of elderly CMM patients, so that patients can benefit from this clinical tool through early detection and timely intervention. The method includes the following steps:

[0040] The data of CHARLS database in 2011 and 2015 were selected for analysis in this invention. After excluding the participants with missing data, 1424 elderly CMM patients were included in the analysis. They were randomly divided into a training set of 996 people and an internal validation set of 428 people at a ratio of 7:3. In addition, the data of 3335 elderly CMM patients with the final 6 predictive variables in 2018 and 2020 in the CHARLS database were included as an external validation set. The detailed inclusion and exclusion flowchart of the study population is shown in Figure 2.

[0041] The outcome variable mainly of interest in this invention is disability, which is a binary variable:

[0042] 1. No disability: The scores of the Activities of Daily Living (ADL) scale and the Instrumental Activities of Daily Living (IADL) scale = 0 and self-reported as non-disabled;

[0043] 2. Disability: ADL / IADL score ≥ 1 or self-reported as disabled.

[0044] This invention selected 46 factors that may be related to the disability of elderly CMM patients as candidate predictive variables, including:

[0045] 1. Demographic and mental health variables: age, gender (male, female), depression, and life satisfaction;

[0046] 2. Lifestyle and health behavior variables: sleep duration, social activities, smoking (including current smoking and smoking history), and alcohol consumption (including current alcohol consumption and drinking history);

[0047] 3. Laboratory examination variables: Triglyceride-Glucose Index (TyG), Triglyceride-Glucose-Body Mass Index (TyG-BMI), Chinese Visceral Adiposity Index (CVAI), Fasting Blood Glucose (FBG), Glycated Hemoglobin (HbA1c), Total Cholesterol (TC), Triglyceride (TG), Low-Density Lipoprotein Cholesterol (LDL-C), High-Density Lipoprotein Cholesterol (HDL-C), White Blood Cell (WBC), C-Reactive Protein (CRP), Platelet (PLT), Mean Corpuscular Volume (MCV), Hemoglobin (HGB), Hematocrit (HCT), Creatinine (CREA), Blood Urea Nitrogen (BUN), Cystatin C (CYsC), Uric Acid (UA);

[0048] 4. Physical examination variables: mean pulse, maximum left hand grip strength, maximum right hand grip strength, mean walking speed, Body Mass Index (BMI), and waist circumference;

[0049] 5. Clinically relevant variables: chest pain symptoms, cognitive function, falls, hip fractures, hypertension, heart disease, stroke, diabetes, kidney disease, dyslipidemia, and the number of CMMs; among which the binary variables are: gender, smoking, alcohol consumption, falls, hip fractures, hypertension, heart disease, stroke, diabetes, kidney disease, dyslipidemia, and the rest are numerical variables.

[0050] Data collection methods:

[0051] 1. Demographic and mental health variables, lifestyle and health behavior variables, clinically relevant variables, outcome variables: These four related variables were obtained through questionnaires by trained staff. Among them, depression was evaluated using the Center for Epidemiologic Studies Depression Scale (CES-D-10), social participation was obtained by calculating the total number of social activities participated by the participants, and cognitive function was evaluated using mental status and episodic memory.

[0052] 2. Laboratory test variables: First, a complete blood count was performed immediately after sample collection at the local county-level health center. Subsequently, the samples were transported back to the research headquarters for analysis of other biomarkers. The calculation formulas for TYG, TYG-BMI, and CVAI are as follows:

[0053] TyG = Ln[1 / (2 × fasting blood glucose (mg / dL) × fasting triglycerides (mg / dL))];

[0054] TyG-BMI: BMI = weight (kg) / height (m 2 )

[0055] TyG-BMI = TyG × BMI;

[0056] For men: CVAI = -267.93 + 0.68 × age (years) + 0.03 × BMI (kg / m²) + 4.00 × WC (cm) + 22.00 × log10(TG) (mmol / L) - 16.32 × HDL-C (mmol / L);

[0057] For women: CVAI = -187.32 + 1.71 × age (years) + 4.23 × BMI (kg / m²) + 1.12 × WC (cm) + 39.76 × log10(TG) (mmol / L) - 11.66 × HDL-C (mmol / L).

[0058] 3. Physical examination variables: The physical examinations of the participants were obtained by the measuring personnel using professional equipment. Among them, the average pulse was calculated using the pulses measured in the 2nd and 3rd measurements, the maximum grip strength of both the left and right hands was measured using a dynamometer, the walking speed was measured twice using a stopwatch and the average value was taken, the body mass index was calculated based on height and weight, and the waist circumference was measured using a flexible tape measure.

[0059] The statistical analysis methods are as follows: All numerical variables are expressed as the median and interquartile range (non-normal distribution), and the Wilcoxon rank sum test is used for analysis of between-group comparisons. Categorical variables are expressed as percentages, and the χ2 test or Fisher's exact test is used for analysis of between-group comparisons. Odds ratio (OR) and 95% confidence interval (CI) are used as effect estimators. R software is used for all analyses in this study. All tests are two-tailed tests, and P < 0.05 is considered statistically significant.

[0060] Among the 1424 subjects, 828 (58.1%) were evaluated as having disability symptoms, which further verified the high incidence of disability in elderly CMM patients, as shown in the between-group comparison of the disabled group and the non-disabled group in Table 1.

[0061] Sixteen variables were initially screened according to univariate Logistic regression analysis in the training set data, and the results of univariate Logistic regression analysis are shown in Table 2.

[0062] Nine significantly correlated variables were further screened by Lasso regression analysis.

[0063] Specifically, Lasso regression analysis based on 10-fold cross-validation: As the value of the regularization parameter Lambda increases, the regression coefficients of each variable gradually tend to zero, and the number of non-zero coefficient variables also continuously decreases (see Figure 3). Vertical lines were drawn at the minimum value of λ (λ = 0.0059, Logλ = -5.1378) and 1 standard error of the minimum value of λ (λ = 0.0286, Logλ = -3.5543) respectively (see Figure 4). Finally, we selected 1 standard error of the minimum value of λ as the optimal value and screened out 9 non-zero coefficient variables, including depression, cognitive function, stroke, maximum grip strength of the left hand, maximum grip strength of the right hand, number of CMMs, age, falls, and social activities.

[0064] The 9 screened variables were then subjected to multivariate Logistic regression analysis, and 6 variables with p value < 0.05 were included in the nomogram prediction model as the final prediction variables, as shown in Table 3.

[0065] Table 1: Baseline characteristics of the study population

[0066]

[0067]

[0068]

[0069] Note: Variables marked with * are binary classification variables. HCT is hematocrit, FBG is fasting blood glucose, HbA1c is glycated hemoglobin, PLT is platelet, HGB is hemoglobin, TC is total cholesterol, HDL-C is high-density lipoprotein cholesterol, CREA is creatinine, MCV is mean corpuscular volume, WBC is white blood cell, TG is triglyceride, BUN is blood urea nitrogen, CRP is C-reactive protein, CVAI is Chinese visceral adiposity index, LDL-C is low-density lipoprotein cholesterol, UA is uric acid, CysC is cystatin C, BMI is body mass index, TyG is triglyceride-glucose index, and TyG-BMI is triglyceride-glucose-body mass index

[0070] Table 2: Results of univariate Logistic regression analysis

[0071]

[0072] Table 3: Results of multivariate Logistic regression analysis

[0073]

[0074] The nomogram prediction model is shown in Figure 5. For each variable, draw a vertical line to the scoring axis to obtain the score, and finally add up the scores to get the total score. The vertical line corresponding to the total score downward represents the disability probability of elderly CMM patients; the higher the total score, the greater the disability risk of elderly CMM patients. The corresponding scores for each variable are as follows:

[0075] Age (60, 68, 76, 84) corresponds to scores (31, 40, 49, 59);

[0076] Cognitive function (0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22) corresponds to scores (89, 80, 72, 63, 55, 46, 38, 30, 21, 13, 4, -4);

[0077] Falls (no falls, falls) corresponds to scores (40, 53);

[0078] Depression (0, 5, 10, 15, 20, 25, 30) corresponds to scores (14, 28, 43, 57, 71, 86, 100);

[0079] Number of CMMs (2, 3, 4, 5) corresponds to scores (35, 44, 53, 62);

[0080] Stroke (no stroke, stroke) corresponds to scores (40, 69);

[0081] Total score corresponds to the predicted probability:

[0082] The predicted probability for 100 points is 1.6%; for 150 points is 7.4%; for 200 points is 27.9%; for 250 points is 65.4%; for 300 points is 90.2%; for 350 points is 97.8%; for 400 points is 99.5%.

[0083] The dynamic nomogram prediction model is shown in Figures 6 and 7. This prediction model is developed for the web based on the shiny package in R software. The shiny package is a web development framework. The shinyPredict package is used to apply shiny for prediction and draw the dynamic nomogram. shinyapps.io is a cloud platform for managing shiny web applications, which can deploy and host shiny applications to the cloud.

[0084] Demonstration of the application of the nomogram and the dynamic nomogram:

[0085] As shown in Figure 5: Age = 85; Depression score = 8; Cognitive function score = 9.5; Stroke (no); Number of CMMs = 2; Fall (no); The finally predicted probability of disability in elderly CMM patients = 72.1%.

[0086] As shown in Figures 6 and 7: Age = 68; Depression score = 0; Cognitive function score = 12; Stroke (yes); Number of CMMs = 3; Fall (yes); The finally predicted probability of disability in elderly CMM patients = 71.1%.

[0087] In step S4:

[0088] The discrimination of the model is evaluated through the ROC curve. The areas under the curves (Area Under Curve, AUC) of the training set (Figure 8), internal validation set (Figure 9), and external validation set (Figure 10) are 0.746 (95% CI: 0.7157 - 0.7766), 0.740 (95% CI: 0.6928 - 0.7874), and 0.754 (95% CI: 0.7373 - 0.7697), respectively. These data indicate that the nomogram model has good discrimination ability and can correctly identify disabled and non-disabled patients.

[0089] The calibration of the model was evaluated by Hosmer-Lemeshow goodness-of-fit test and calibration curve (Bootstrap method, n = 1000). The results showed that for the training set (Figure 11), internal validation set (Figure 12), and external validation set (Figure 13), the P values were 0.9273, 0.373, and 0.7675, respectively (all P values > 0.05), and the Brier scores were 0.199, 0.204, and 0.201, respectively (all Brier scores < 0.25). These data indicated that the model had good consistency and the predicted disability probability was highly consistent with the actual disability probability.

[0090] The clinical application value of the model was evaluated by DCA curve. DCA showed that the net benefit of the prediction model in the training set (Figure 14), internal validation set (Figure 15), and external validation set (Figure 16) was significantly higher than the two extreme cases, indicating that the model had significant net benefit and prediction accuracy.

[0091] In summary, the construction method and application of the CMM elderly patient disability risk prediction model are as follows:

[0092] Based on independent influencing factors, a nomogram prediction model was constructed. At the same time, the online dynamic nomogram can more conveniently and quickly predict the risk probability of disability in elderly CMM patients, realizing the early detection, diagnosis, and treatment of disabled patients, thereby improving the short-term and long-term prognosis of elderly CMM patients.

[0093] Through the above specific implementation manners, those skilled in the art of the present invention can easily implement the present invention. However, it should be understood that the present invention is not limited to the above specific implementation manners. Based on the disclosed implementation manners, those skilled in the art can arbitrarily combine different technical features to implement different technical solutions.

Claims

1. A method for constructing a disability risk prediction model for elderly patients with cardiovascular and metabolic comorbidities and its application, characterized in that: The following steps are involved: S1. Data collection and screening: The CHARLS database was used to collect and screen data on elderly patients with CMM, including disability risk factors and outcome indicators; S2. Predictor variable screening: Single-factor logistic regression was followed by further Lasso regression and multi-factor logistic regression analysis to screen predictor variables and construct a nomogram model; S3. Prediction model development: Based on the selected predictive variables, the dynamic nomogram prediction model was constructed using the shinyPredict package; S4. Internal and external validation and evaluation of the prediction model; The results of the multivariate logistic regression analysis described in step S2 showed that age, OR=1.03, 95%CI: 1.00-1.06, P=0.021; number of CMMs, OR=1.32, 95%CI: 1.06-1.64, P=0.012; depression, OR=1.09, 95%CI: 1.06-1.12, P<0.001; falls, OR=1.47, 95%CI: 1.02-2.15, P=0.041; stroke, OR=2.55, 95%CI: 1.55-4.34, P<0.001 were independent risk factors for disability; cognitive function, OR=0.89, 95%CI: 0.85-0.93, P<0.001 was a protective factor for disability.

2. The method for constructing a CMM elderly patient disability risk prediction model based on a dynamic nomogram according to claim 1, characterized in that: In step S1, CMM is defined as an individual suffering from two or more of the following diseases: hypertension, heart disease, stroke, diabetes, and dyslipidemia. Heart disease is defined as a general term for a variety of heart-related diseases, including myocardial infarction, coronary heart disease, angina pectoris, congestive heart failure, etc. The above diseases are all binary variables based on patient self-report and confirmed by doctor's diagnosis.

3. The method for constructing a CMM elderly patient disability risk prediction model based on a dynamic nomogram according to claim 2, characterized in that: Data collection and screening described in step S1: The present invention selected data from the CHARLS database 2011 and 2015 for analysis. After excluding participants with missing data, 1,424 elderly patients with CMM were included in the analysis and randomly divided into a training set of 996 people and an internal validation set of 428 people in a ratio of 7:3 based on R software. In addition, the present invention also included 3,335 elderly patients with CMM in 2018 and 2020 in the CHARLS database containing the final 6 predictor variables as an external validation set.

4. The method for constructing a CMM elderly patient disability risk prediction model based on a dynamic nomogram according to claim 3, characterized in that: The disability risk factors described in step S1 include demographic and mental health variables, lifestyle and health behavior variables, laboratory test variables, physical examination variables, and clinically related variables.

5. The method for constructing a CMM elderly patient disability risk prediction model based on a dynamic nomogram according to claim 4, characterized in that: The outcome indicator described in step S1 was disability, which was assessed by ADL and IADL. Those with ADL and IADL scores = 0 and self-reported as non-disabled were considered to have no disability, and those with ADL / IADL scores ≥ 1 or self-reported as disabled were considered to have disability.

6. The method for constructing a CMM elderly patient disability risk prediction model based on a dynamic nomogram according to claim 5, characterized in that: Screening of predictive variables described in step S2: In the training set data, 16 predictive variables were first preliminarily screened out through univariate logistic regression analysis; then, lasso regression analysis was used to further screen the variables, and the optimal value of the regularization parameter (λ) was determined through 10-fold cross-validation, and finally 9 relevant predictive variables were screened out; finally, multivariate logistic regression analysis was performed on these 9 variables, and 6 predictive variables with a P value less than 0.05 were included in the nomogram prediction model.

7. The method for constructing a CMM elderly patient disability risk prediction model based on a dynamic nomogram according to claim 6, characterized in that: In step S3, the dynamic nomogram prediction model constructed based on the shinyPredict package of R software can calculate the probability of disability of elderly patients with CMM according to the input variables.

8. The method for constructing a CMM elderly patient disability risk prediction model based on a dynamic nomogram according to claim 7, characterized in that: In step S4, internal validation was performed using the bootstrap method, and external validation was performed using data from different time periods, i.e., the data of 3,335 elderly patients with CMM in 2018 and 2020 containing the final 6 predictor variables were used to complete the external validation; the model was evaluated as follows: ROC curve, calibration curve and DCA curve were used to respectively evaluate the discrimination, calibration and clinical utility of the prediction model.

9. Obtain a CMM elderly patient disability risk prediction model based on the method for constructing a CMM elderly patient disability risk prediction model based on a dynamic nomogram as described in any one of claims 1 to 8.

Images