TWI854592B - Methods and systems for predicting risk of cardiovascular disease - Google Patents

Abstract

The present invention provides methods and computer-implemented methods which are used to predict a risk of cardiovascular disease by using a machine learning mode to analyze a relationship between the occurrence of cardiovascular disease and the health data of patients.

Description

Method and system for predicting the risk of cardiovascular disease

The present invention relates to a method and a computer-implemented method for predicting the risk of cardiovascular disease by using a machine learning model to analyze the relationship between the occurrence of cardiovascular disease and the patient's health data.

High blood pressure is one of the most common public health problems in the world. High blood pressure increases the risk of serious health problems, such as stroke, kidney failure, and cardiovascular disease (CVD). If high blood pressure is not treated promptly, it can lead to a heart attack or stroke. In recent years, researchers have gained a deeper understanding of the pathophysiology of high blood pressure and developed more effective treatments and prevention methods. Among them, home blood pressure (HBP) levels have been shown to predict future high blood pressure and cardiovascular disease.

Poor blood pressure control is influenced by multiple factors, including obesity, salt sensitivity, psychological stress, genetic predisposition, epigenetics (regulation of gene expression without changing the DNA sequence), sleep apnea, autonomic regulation, microbiome, and environmental factors such as smoking, excessive salt intake, excessive alcohol consumption, and socioeconomic status. Often, these factors interact in complex ways and affect various diseases. Because of this, traditional statistical models rarely reflect all the complex causal relationships between risk factors. Therefore, comprehensive data analysis is essential to develop accurate disease prediction models. In the past, various statistical methods have been used to develop prediction models and discover important risk factors. However, artificial intelligence (AI) and big data have emerged recently and are increasingly being used to develop disease prediction models.

The present invention relates to methods and computer-implemented methods for assessing an individual's risk of developing a cardiovascular disease event within a period of 1 month to 4 years. These methods use machine learning models to analyze the relationship between the occurrence of cardiovascular disease and the individual's health data.

CVD prediction is an important goal of modern medicine. Predicting CVD risk to enable early treatment can significantly reduce the risk of serious illness. Blood pressure (BP) is a key indicator measured in most health examinations. In the past, BP could only be measured in hospitals or health centers; however, these measurements are now more accessible if BP monitoring can be performed regularly at home using lightweight BP monitoring devices. According to the results shown in this article, the relationship between CVD and BP is clearly revealed. Daily BP monitoring using wearable devices to record exercise volume, alcohol intake, medication dosage, and other data can help to treat cardiovascular diseases in a timely and accurate manner in the future.

If the patient's blood pressure data before CVD occurs can be obtained, the prediction of CVD can also be improved.

As used herein, the terms "a" or "an" are used to describe elements and components of the present invention. This is done only for convenience and to give the basic concept of the present invention. Further, the description should be understood to include one or at least one, and unless the context clearly indicates otherwise, singular terms include the plural and plural terms include the singular. When used in conjunction with the word "comprising" in the scope of the patent application, the terms "a" or "an" can mean one or more.

The term "or" as used herein may mean "and/or".

The present invention provides a method for generating a prediction model for estimating cardiovascular disease (CVD) risk, comprising: (a) obtaining a database from one or more sources, wherein the database comprises health data of non-CVD patients and CVD patients; and data related to the CVD onset time of the CVD patients, and the health data comprises demographic data, personal habit data, disease data, treatment data, blood analysis data and blood pressure data; (b) converting the database into a prediction model for estimating cardiovascular disease (CVD) risk; Input into at least one machine learning model to train the at least one machine learning model to predict the occurrence of CVD; (c) evaluate the accuracy of the at least one machine learning model from the step (b), and when the accuracy of a first machine learning model is higher than an accuracy threshold, select the first machine learning model from the at least one machine learning model; and (d) use the first machine learning model to generate the prediction model for estimating the CVD risk at different time points.

In one specific embodiment, the one or more sources include the database of the Taiwan Hypertension-Related Heart Disease Clinical Trial Consortium and the database of the Taiwan Center for Health and Welfare Data Science.

In certain aspects, the non-CVD patient is a patient initially diagnosed as not having CVD, and the CDV patient is a patient initially diagnosed as having CVD.

In another embodiment, the CVD comprises myocardial infarction, stroke, heart failure, cardiovascular death, or a combination thereof.

In one embodiment, the demographic data includes age, gender, body mass index (BMI), and waist circumference.

In another specific embodiment, the personal habit data includes smoking habits, drinking habits, and exercise habits. In some aspects, smoking habits and drinking habits are collected once a month or a year, and exercise habits are collected once a week.

In a specific embodiment, the disease data includes hypertension (HT), diabetes mellitus (DM), hyperlipidemia (HL) or a combination thereof.

In another specific embodiment, the treatment data includes the use of antihypertensive drugs, antihyperglycemic drugs, antihyperlipidemic drugs, aspirin or a combination thereof.

In a specific embodiment, the blood analysis data includes aspartate aminotransferase (GOT), alanine aminotransferase (GPT), blood glucose, glycosylated hemoglobin (such as HbA1c), cholesterol, low-density lipoprotein (LDL), high-density lipoprotein (HDL) or a combination thereof.

In a specific embodiment, the blood pressure data include data of systolic blood pressure (SBP) and data of diastolic blood pressure (DBP). In a preferred specific embodiment, the SBP data include the mean value of SBP, the standard deviation (SD) of SBP, the coefficient of variation (CV) of SBP, the average real variability (ARV) of SBP, the maximum and minimum values of the morning and evening average (MEave) of SBP over a period of time, or a combination thereof. In a more preferred specific embodiment, the SBP data include the maximum and minimum values of the morning and evening average of SBP over a period of time. In another specific embodiment, the DBP data include the mean value of DBP, the SD of DBP, the CV of DBP, the ARV of DBP, the maximum and minimum values of MEave of DBP over a period of time, or a combination thereof. In some aspects, the MEave of SBP and DBP is collected by obtaining blood pressure before eating at 6 am to 8 am/4 pm to 6 pm. In another embodiment, the collection period of SBP and DBP is 1 to 12 weeks. In a preferred embodiment, the collection period of SBP and DBP is 1 to 6 weeks. In a more preferred embodiment, the collection period of SBP and DBP is 1 week.

In one embodiment, the database further includes temperature data and air pollution data. In a preferred embodiment, the air pollution data includes air quality index (AQI), O ₃ , PM10 and PM2.5. In a more preferred embodiment, the air pollution data includes PM2.5.

In the present invention, the database is randomly divided into a training set and a validation set for training the at least one machine learning model. Usually, the training set is used together with the validation set. The term "validation set" refers to a group of patients in a statistical sample, where the data of the individuals are used to validate or evaluate the quantitative value of the training set determined to be used.

In one embodiment, the at least one machine learning model includes logistic regression (LR), deep neural networks (DNNs), random forest (RF), light gradient boosting machine (LightGBM), eXtreme gradient boosting (XGboost), decision trees (DTs), k nearest neighbor (KNN), Adaboost, gradient boosting (Gboost), DT bagging (DTB), knn bagging (KNNB) or RF bagging (RFB). In a preferred embodiment, the at least one machine learning model includes XGBoost, DTB or RF. In a preferred embodiment, the first machine learning model includes XGBoost, DTB or RF.

In addition, the at least one machine learning model includes a deep-learning architecture. In some aspects, the deep-learning architecture includes deep neural networks, deep belief networks, deep reinforcement learning, recurrent neural networks, convolutional neural networks, and transformers.

In the present invention, the at least one machine learning model may be a supervised machine learning algorithm. The supervised machine learning algorithm may be trained by using the patient's previous data, previous data of similar patients, or a combination thereof. The supervised machine learning algorithm may be a regression algorithm, a support vector machine, a decision tree, a neural network, etc. In the case where the machine learning model is a regression algorithm, the weights may be regression parameters. The supervised machine learning algorithm may be a binary classifier that predicts whether a patient will experience a CVD event. The binary classifier may generate a probability risk score between 0 and 1. In some cases, the system can map the probabilistic risk scores to qualitative risk categories. Alternatively, the supervised machine learning algorithm can be a multivariate classifier that directly predicts the qualitative risk category. In addition, internal validation of the at least one machine learning model is performed using ten-fold cross validation.

In another specific embodiment, the threshold of the accuracy is 0.85. In a preferred specific embodiment, the threshold of the accuracy is 0.9. In a more preferred specific embodiment, the threshold of the accuracy is 0.95.

In a specific embodiment, the data related to the onset time of CVD in the CVD patient includes the onset time of a CVD event in the CVD patient within 1 month, 3 months, 6 months, 9 months, 1 year, 2 years, 3 years or 4 years. In certain aspects, the at least one machine learning model is trained to analyze the relationship between the health data of the non-CVD patient and the CVD patient and the data related to the onset time of CVD in the CVD patient to predict the occurrence of future CVD. Therefore, the present invention can use data related to different CVD onset times of CVD patients to train the at least one machine learning model to generate different prediction models, and the different prediction models are used to predict the probability of a CVD event occurring in an individual at different time points in the future, wherein the different time points are within 1 month, 3 months, 6 months, 9 months, 1 year, 2 years, 3 years or 4 years.

In the present invention, the patient's health data is used to establish a high-accuracy prediction model for the occurrence of long-term (1 to 4 years) CVD. In addition, the patient's health data, the temperature data and the air pollution data are used to establish a high-accuracy prediction model for the occurrence of short-term (1 to 9 months) CVD. Therefore, the prediction model of the present invention can estimate the risk of CVD at different time points. In a specific embodiment, the time point includes 1 month, 3 months, 6 months, 9 months, 1 year, 2 years, 3 years or 4 years.

The present invention also provides a system, which includes one or more computers and one or more storage devices for storing an executable program. When the one or more computers are run, the one or more computers are caused to perform the following operations to generate a prediction model for estimating CVD risk, including: (a) obtaining a database from one or more sources, wherein the database includes health data of non-CVD patients and CVD patients; and data related to the CVD onset time of the CVD patients, and the health data includes demographic data, personal habit data, disease data, data, treatment data, blood analysis data and blood pressure data; (b) inputting the database into at least one machine learning model to train the at least one machine learning model to predict the occurrence of CVD; (c) evaluating the accuracy of the at least one machine learning model from the step (b), and when the accuracy of a first machine learning model is higher than an accuracy threshold, selecting the first machine learning model from the at least one machine learning model; and (d) using the first machine learning model to generate the prediction model for estimating the CVD risk at different time points.

The present invention further provides a method for predicting an individual's CVD risk, comprising: (i) obtaining health data of the individual, wherein the health data includes demographic data, personal habit data, disease data, treatment data, blood analysis data and blood pressure data; (ii) inputting the health data into the above-mentioned prediction model for estimating CVD risk; and (iii) outputting CVD risk prediction results at different time points.

In another specific embodiment, the prediction result of the prediction model for estimating CVD risk provides the probability of the individual having a CVD event within 10 years. In a preferred specific embodiment, the prediction result of the prediction model for estimating CVD risk provides the probability of the individual having a CVD event within 6 years. In a more preferred specific embodiment, the prediction result of the prediction model for estimating CVD risk provides the probability of the individual having a CVD event within 4 years.

In a specific embodiment, the prediction result of the prediction model for estimating CVD risk provides the probability of the individual experiencing a CVD event within 1 month, 3 months, 6 months, 9 months, 1 year, 2 years, 3 years or 4 years.

In another specific embodiment, the method further comprises a step (iv) after step (iii), wherein the step (iv) comprises determining whether to initiate medical intervention for the individual at risk of CVD based on the prediction result.

The present invention further provides a computer-implemented method for determining the CVD risk of an individual, comprising: receiving health data of the individual; using the above-mentioned prediction model for estimating CVD risk and determining the CVD risk of the individual based on the received health data; and outputting the CVD risk determined at different time points, wherein the prediction model for estimating CVD risk evaluates the determined CVD risk by analyzing the relationship between the occurrence of CVD and the received health data.

In one embodiment, the individual and the patient are human.

The present invention also provides a health care system, which includes: (a) a patient monitoring module for collecting patient data generated by real-time monitoring of the patient; (b) a database for collecting the patient's health data, wherein the health data includes demographic data, personal habit data, disease data, treatment data, blood analysis data and blood pressure data; and (c) an integration module for receiving the patient data and the health data, and using the above-mentioned prediction model for estimating CVD risk to analyze the patient data and the health data, and output the prediction results of the patient's CVD risk at different time points, which are based on the analysis of the prediction model for estimating CVD risk. In the health care system, the patient monitoring module is a remote patient monitoring module.

In one embodiment, the patient monitoring module is a remote patient monitoring module. Therefore, the remote patient monitoring module is capable of collecting patient data by remotely monitoring the patient in real time. In some aspects, the function of the patient monitoring module includes capturing vital signs. In addition, the purpose of the remote patient monitoring module is to monitor the patient's home blood pressure.

In another specific embodiment, the patient monitoring module, the database and the integration module are interconnected.

10:Health care system

101: Patient monitoring module

102: Database

103: Integration module

Figure 1 is a description of the data source of the present invention. n: number of patients.

Figure 2 shows a healthcare system used to estimate a patient's risk of cardiovascular disease.

The embodiments of the present invention may have different implementation contents and are not limited to the examples given below. The following embodiments only represent various aspects and features of the present invention.

1. Research design

The model of the present invention is constructed from two databases. One is collected from the Taiwan Consortium of Hypertension-associated Cardiac Disease (TCHC) database, which has 11 participating medical centers and is a non-profit research consortium focusing on clinical trials and research collaborations in hypertension and hypertension-related diseases; the other is the Taiwan Health and Welfare Data Science Center (HWDSC) database, which contains records of Taiwan's National Health Insurance beneficiaries, outpatient visits, hospitalizations, pharmacy databases, and cause of death databases from 2000 to 2017. The combined database has nearly 2,820 participants (Figure 1); approximately 80% and 20% of the datasets were randomly selected for training and validation, respectively. Since only 20% of the patients in the dataset developed CVD, 80% of the CVD-positive and 80% of the CVD-negative datasets were used for training. The remaining 20% of the items were used for testing. TCHC contains cardiovascular disease records, blood pressure data, medication records, lifestyle interviews, basic health information, and other disease records. This not only helps to understand the impact of measurable health values on cardiovascular disease, but also helps to understand whether there are interactions between different diseases. Several algorithms were used: Logical Regression (LR), Deep Neural Network (DNN), Random Forest (RF), Lightweight Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (XGboost), Decision Trees (DTs), K-Nearest Neighbors (KNN), Adaptive Boosting (Adaboost), Gradient Boosting (Gboost), DT Bagging (DTB), knn Bagging (KNNB) and RF Bagging (RFB). Finally, the present invention leaves three models with the best performance: XGBoost, DTB and RF.

The ratio of CVD results is shown in Table 1.

Table 1. Proportion of CVD results

2. Data processing

The main features are as follows:

2-1. Characteristics not related to BP:

Demographics: age, gender, BMI, waist circumference

Lifestyle habits: smoking habits, drinking habits, exercise habits

Diseases: Hypertension (HT), diabetes mellitus (DM), hyperlipidemia (HL)

Treatment: antihypertensive drugs (anti-BP drugs), hypoglycemic drugs, lipid-lowering drugs, aspirin

Others: GOT, GPT, blood sugar (Glu_AC and Glu_PC), HbA1c, cholesterol, LDL and HDL

2-2. BP-related characteristics

Systolic blood pressure (SBP): mean, standard deviation (SD), coefficient of variation (CV), average actual variation (ARV), and maximum and minimum values of the 7-day morning and evening mean (MEave) (6am to 8pm/4pm to 6pm before meals).

Diastolic blood pressure (DBP): mean, SD, CV, ARV, and maximum and minimum values of MEave

2.3 Weather

Temperature (If the patient was diagnosed with cardiovascular disease on a particular day, use the temperature of that day. If the patient did not have cardiovascular disease, use the temperature of the day when blood pressure was measured.)

Air pollution values: Includes commonly used air pollution indicators such as AQI, O ₃ , PM10 and PM2.5.

Age, gender, BMI, waist circumference are common features that are easily attached to the medical model. Habits (smoking, drinking, exercise) have an impact on a person's health status. The collected data related to smoking, drinking, and exercise habits are shown in Table 2. As mentioned in the present invention, diseases may affect each other; therefore, the present invention includes hypertension, diabetes, hyperlipidemia, cancer, panic, anger and depression as features. Aspartate aminotransferase (GOT), alanine aminotransferase (GPT) and blood sugar are risk factors for health. Low-density lipoprotein (LDL), sometimes called "bad" cholesterol, and high-density lipoprotein (HDL), or "good" cholesterol, are important parameters of cholesterol. The goal is to predict the occurrence of CVD by using BP data. However, BP data is complex. Therefore, a method using SBP and DBP data is used instead. Some common CVD-related diseases are also analyzed. Age-based regression is used for missing data. Patients with and without CVD in the dataset are labeled 1 and 0, respectively. Similarly, patients with and without HT, DM, HL, and cancer are also labeled 1 and 0, respectively. For numerical data, the original data is used. The present invention uses a feature scaling method to standardize the effect of each feature. For categorical data, one-hot encoding is used to convert the features into trainable ones. Tables 3 and 4 show the continuous variables and binary variables used in the present invention. In addition, the present invention uses recursive feature elimination (RFE) based on logical regression and principal component analysis (PCA) to perform feature selection.

Table 2. Classification of smoking, drinking, and exercise habits

Table 3. Continuous variables

Table 4. Binary variables

Some important features were defined as follows: mean of morning and evening BP values (MEave), SD of MEave for home systolic and diastolic BP, CV of MEave for home systolic and diastolic BP, ARV of MEave for home systolic and diastolic BP, and variability independent of the mean (VIM) of MEave for home systolic and diastolic BP. Despite the inclusion of BP data, features related to CVD were selected for model construction. TCHC includes systolic and diastolic BP for each patient four times daily for 7 days. Interpolation was used to reconstruct missing data. The method accurately identified BP characteristics, variability, and trends.

After experiments, it was found that most blood pressure related parameters do not affect the accuracy of the model. Therefore, in Tables 5, 6 and 7, the present invention only uses the MEave of SBP as a prediction parameter to reduce the training load of the model.

3. Model

Based on previous experiments, the present invention uses three models with the highest accuracy in predicting the occurrence of CVD. When the present invention predicts the occurrence of short-term CVD, the present invention can obtain good results. Otherwise, when predicting the occurrence of long-term CVD, the accuracy will drop to about 0.85. However, the area under the curve (AUC) is mostly very high, close to 0.9.

4. CVD prediction of temperature and air pollution

Tables 5, 6, and 7 show the results obtained by adding air temperature and air pollution to the above model. The temperature is calculated as the average temperature of the 7 days when blood pressure is measured, and the air pollution is calculated as the average PM2.5 of the 7 days when blood pressure is measured. From Table 6, it can be seen that after adding air temperature or air pollution, the accuracy of the model is slightly improved.

For temperature, only the temperature on the day the cardiovascular disease occurred was used because the effect of temperature is immediate.

Table 5. Comparison results of models for short-term CVD with only temperature

As can be seen from Table 5, using temperature alone to predict the occurrence of cardiovascular disease can achieve an accuracy of more than 0.88, which is already a very good prediction model. However, more features can be added to improve the accuracy.

Since the characteristics to be included include those involving basic vital signs, blood pressure characteristics and air pollution, which are not only short-term factors but also common causes of short- to medium-term cardiovascular diseases; the present invention expands the data set to include patients who have experienced cardiovascular diseases within 1, 3, 6 and 9 months.

As can be seen in Table 6, after adding other features, the accuracy of the model has been significantly improved, reaching 0.99, and the average is above 0.95. This can be considered a very accurate short-term prediction model for cardiovascular disease.

Table 6. Comparison results of models for short-term CVD occurrence with temperature and air pollution

5. Long-term model

The short- to medium-term models have shown good results. Next, the present invention wants to try the long-term prediction of cardiovascular disease. The time range set by the present invention is for patients who have experienced cardiovascular disease within 1, 2, 3 and 4 years. As there are more long-term patients and more variables to consider, the accuracy of the present invention is expected to decrease.

As shown in Table 7, except for the accuracy of predicting cardiovascular disease within four years, which is about 0.88, the model's prediction accuracy in 1, 2, and 3 years is all above 0.9. Compared with the medium- and short-term models, although there is a slight gap, it is still a model with good accuracy.

Table 7. Comparison of models for long-term CVD occurrence

The results of the present invention are also validated by categorical variable screening, such as the relationship between body size, exercise, alcohol and drugs and CVD. Moreover, it is not easy for traditional regression models, such as LR, to achieve high accuracy. In contrast, machine learning models are suitable for dealing with biological binary judgment models. There are several models suitable for predicting CVD risk, such as RF, DTB and xGBoost. The purpose of the present invention is to identify CVD risk. However, based on the limited database size, BP records are only 7 days; extended records of 4-6 weeks may produce more accurate results.

In summary, participants in the present invention were divided into short-term group (within 1/3/6/9 months) and long-term group (within 1/2/3/4 years) based on the occurrence of CVD. Depending on the time period, the input factors will also be different.

For short-term models: A combination of basic physical characteristics, blood pressure data, and temperature is used to predict short-term CVD occurrence.

For long-term models: basic physical features are used to predict the occurrence of long-term CVD.

The model constructed by the present invention can be applied to the prediction of short-term acute cardiovascular diseases, and can also be used as a prediction model for long-term follow-up observation and preventive medicine.

In addition, the present invention further provides a health care system for estimating the cardiovascular disease risk of a patient. In Figure 2, the health care system 10 includes: (a) a patient monitoring module 101, which is used to collect patient data generated by real-time monitoring of the patient; (b) a database 102, which is used to collect the patient's health data, wherein the health data includes demographic data, personal habit data, disease data, treatment data, blood analysis data and blood pressure data; and (c) an integration module 103, which is used to receive the patient data and the health data, and use the above-mentioned prediction model for estimating CVD risk to analyze the patient data and the health data, and output the prediction results of the patient's CVD risk at different time points, which are based on the analysis of the prediction model for estimating CVD risk. In the healthcare system, the patient monitoring module is a remote patient monitoring module.

Those skilled in the art will understand the above summaries as descriptions of methods used to communicate deposited application information. Those skilled in the art will recognize that these are merely illustrative in nature and that many equivalents are possible.

Claims (12)

A method for generating a prediction model for estimating cardiovascular disease (CVD) risk, comprising: (a) obtaining a database from one or more sources, wherein the database comprises health data, temperature data and air pollution data of non-CVD patients and CVD patients; and data related to the CVD onset time of the CVD patients, and the health data comprises demographic data, personal habit data, disease data, treatment data, blood analysis data and blood pressure data, wherein the blood pressure data comprises the data of systolic blood pressure (SBP) and the data of diastolic blood pressure (DBP), and the data of SBP and DBP comprises the mean value of SBP and DBP, the standard deviation of SBP and DBP, the average ... (a) inputting the database into at least one machine learning model to train the at least one machine learning model to predict the occurrence of CVD; (c) evaluating the accuracy of the at least one machine learning model from the step (b), and when the accuracy of a first machine learning model is higher than an accuracy threshold, selecting the first machine learning model from the at least one machine learning model; and (d) using the first machine learning model to generate the prediction model for estimating the risk of CVD in a time period, wherein the time period is 1 month, 3 months, 6 months or 9 months. The method of claim 1, wherein the CVD comprises myocardial infarction, stroke, heart failure, cardiovascular death or a combination thereof. The method of claim 1, wherein the demographic data includes age, gender, body mass index (BMI) and waist circumference. The method as claimed in claim 1, wherein the personal habit data includes smoking habits, drinking habits, exercise habits or a combination thereof. The method as claimed in claim 1, wherein the disease data includes hypertension, diabetes, hyperlipidemia or a combination thereof. The method as described in claim 1, wherein the treatment data includes the use of antihypertensive drugs, antihyperglycemic drugs, antihyperlipidemic drugs, aspirin or a combination thereof. As described in claim 1, the blood analysis data includes aspartate aminotransferase (GOT), alanine aminotransferase (GPT), blood sugar, glycosylated hemoglobin, cholesterol, low-density lipoprotein (LDL), high-density lipoprotein (HDL) or a combination thereof. The method of claim 1, wherein the at least one machine learning model comprises extreme gradient boosting (XGboost), decision trees bagging (DTB) or random forest (RF). A method for predicting the risk of cardiovascular disease (CVD) of an individual comprises: (i) obtaining health data, temperature data and air pollution data of the individual, wherein the health data comprises demographic data, personal habit data, disease data, treatment data, blood analysis data and blood pressure data, wherein the blood pressure data comprises systolic blood pressure (SBP) data and diastolic blood pressure (DBP) data, and the SBP and DBP data comprises the average value of SBP and DBP, ... (i) inputting the health data, the temperature data and the air pollution data into the prediction model for estimating CVD risk generated by the method of claim 1; and (iii) outputting the prediction result of CVD risk of the individual in a time period, wherein the time period is 1 month, 3 months, 6 months or 9 months. The method of claim 9 further comprises a step (iv) after step (iii), wherein step (iv) comprises determining whether to initiate medical intervention for the individual at risk of CVD based on the prediction result. A health care system, comprising: (a) a patient monitoring module for collecting patient data generated by real-time monitoring of the patient; (b) a database for collecting the patient's health data, temperature data and air pollution data, wherein the health data comprises demographic data, personal habit data, disease data, treatment data, blood analysis data and blood pressure data, wherein the blood pressure data comprises systolic blood pressure (SBP) data and diastolic blood pressure (DBP) data, and the SBP and DBP data comprises the mean value of SBP and DBP, the standard deviation of SBP and DBP, the coefficient of variation ... The invention relates to an integrated module for receiving the patient data, the health data, the temperature data and the air pollution data, and using the prediction model for estimating cardiovascular disease (CVD) risk generated by the method of claim 1 to analyze the patient data, the health data, the temperature data and the air pollution data, and outputting the prediction result of the patient's CVD risk in a time period, which is based on the analysis of the prediction model for estimating CVD risk, wherein the time period is 1 month, 3 months, 6 months or 9 months. A healthcare system as described in claim 11, wherein the patient monitoring module is a remote patient monitoring module.

Images