HyperAI超神经

According to the 2017 diabetes survey, there are 78.13 million elderly people with diabetes in my country. Combined with a number of large-scale population studies, it was found that abnormal glucose metabolism and cardiovascular disease have a high degree of "comorbidity" relationship, that is, diabetic patients often have complications such as coronary heart disease, and the latter has become a major cause of death in diabetic patients - about 75% of diabetic patients died of coronary heart disease. However,Currently, there are only a few studies on the risk factors for survival in patients with coronary heart disease and diabetes or impaired glucose tolerance.

｜Remarks:Impaired glucose tolerance (IGT) is an abnormal glucose metabolism state that transitions from normal blood sugar to diabetes. It is a prediabetes state and may further develop into diabetes mellitus (DM).

In order to break this situation, researchers from the People's Hospital of Macheng City, Hubei Province, China, pioneered a comparison between the logistic regression model (LR) and three machine learning models, successfully predicting the one-year mortality rate of elderly Chinese patients with coronary heart disease combined with diabetes or impaired glucose tolerance, helping the medical community to promptly identify patients at risk of short-term death, thereby providing early warning and treatment.

The study has been published in the journal Cardiovascular Diabetology, titled "Machine learning-based models to predict one-year mortality among Chinese older patients with coronary artery disease combined with impaired glucose tolerance or diabetes mellitus."

Figure 1: This research result has been published in Cardiovascular Diabetology

Paper address:

https://cardiab.biomedcentral.com/articles/10.1186/s12933-023-01854-z

Experimental procedures

Dataset: Data of 451 elderly patients with coronary heart disease from 301 hospitals

This study analyzed 974 elderly patients with coronary heart disease who were admitted to the Department of Geriatric Cardiology of the General Hospital of the Chinese People's Liberation Army from October 2007 to July 2011.The researchers further screened according to two criteria:They are:

1. Aged over 60;

2. Suffering from impaired glucose tolerance (IGT) or diabetes mellitus (DM).

The final dataset contained 451 patients, who were randomly divided into training set (n = 308) and test set (n = 143) in a 7:3 ratio.The training set is used to train and optimize the logistic regression model and three machine learning models, and the test set is used to test the model prediction performance. The data set screening process is as follows:

Figure 2: Flowchart outlining patient selection and study design

Model development: Select 4 major models for horizontal comparison

In this study, the researchers developed a logistic regression model and three machine learning models.The prediction models are established for the gradient boosting machine model (GBM), random forest model (RF) and decision tree model (DT).The prediction effect is evaluated based on several indicators such as Brier Score, AUC (Area Under the Curve), calibration curve and decision curve.

Brier Score:A way to measure the difference between the probability predicted by the algorithm and the actual result. Its value range is 0 to 1, and the higher the score, the worse the prediction result and the lower the degree of calibration.

Figure 3: Brill score calculation formula

AUC:Refers to the area under the curve. In statistics and machine learning, AUC is often used to evaluate the performance of binary classification models. Its value range is from 0 to 1. The closer the value is to 1, the better the model performance is; the closer the value is to 0.5, the weaker the model prediction ability is.

Feature screening and parameter tuning for 3 machine learning models

At the same time, the researchers performed feature screening and parameter tuning on the developed machine learning model.First, they used the LASSO (least absolute shrinkage and selection operator) algorithm combined with 10-fold cross-validation to screen out 7 features that were significantly associated with one-year mortality as model input. These 7 features were hemoglobin, HDL-C, albumin, serum creatinine, NT-proBNP, CHF, and statins. Then, they used random hyperparameter search, 5-fold cross-validation, and bootstrap to find the best parameter combination and obtain the best area under the curve (AUC).

Figure 4: Hyperparameter tuning process

A:Least Absolute Shrinkage and Selection Operator (LASSO) Coefficient Curve

B:The best parameter combination

C:Correlation coefficients between clinical characteristics

From Figure 4, all correlation coefficients are lower than 0.80, indicating that there is no serious collinearity.The above 7 clinical features were used to train the logistic regression model and 3 machine learning prediction models.After model training and optimization, the optimal hyperparameters for each model are shown in the following table:

Table 1: Optimal hyperparameters for each model

Experimental Results

From the overall performance of each model:

* The Brier score of the logistic regression model (LR) is 0.116

* The Brier score of the Gradient Boosting Machine model (GBM) is 0.114

* The Brier score of the decision tree model (DT) is 0.143

* The Brier score of the Random Forest model (RF) is 0.126

The following figure shows the analysis results of each model:

Figure 5: AUC, calibration curve, decision curve, and SHAP value of each model

D:Overall performance of each model

E:Calibration curves for each model

F:Decision curves for each model

G:SHAP value heat map

H:Feature Importance Analysis Based on SHAP

According to Figure 5, the following conclusions can be drawn:

1. The AUCs of LR, GBM, DT and RF models are 0.827, 0.836, 0.760 and 0.829 respectively.

2. The calibration curves show that all models have good calibration effects, among which the GBM model has the best effect.

3. Decision curve analysis showed that both the GBM model and the LR model had good clinical practicality.

4. Based on the GBM model, the researchers further analyzed the importance of significant clinical characteristics in the entire population. By analyzing both individual SHAP values and average SHAP values, they found that the top three characteristics associated with one-year mortality were NT-proBNP, albumin, and statins.

| SHAPE: Shaley Additive exPlanation, feature contribution. By analyzing the SHAP value, researchers can obtain an explanation for the prediction results, understand how each feature affects the model's prediction, and thus better understand and explain the model's behavior.

In summary, the researchers suggested that although the models in previous studies had high predictive performance, they were not suitable for clinical application due to too many variables. In this study, the researchers successfully used 7 features to develop a model to predict one-year mortality.The results show that the GBM model has an AUC of up to 0.836 and a Brier score of 0.116, with the best overall prediction performance.

It is worth noting that in order to further facilitate clinical applications, the researchers also designed an online application that only requires doctors to fill in patient parameters to predict the probability of death within one year. In this way, doctors can take favorable measures for high-risk patients as early as possible to increase the patient's survival probability.

AI medical field has a bright future, but we should not be blindly optimistic

With the gradual maturity of AI voice interaction, computer vision, cognitive computing, deep learning and other technologies, the application scenarios of AI in the medical field are becoming increasingly diverse.It involves multiple directions such as medical imaging, virtual assistants, drug development, health management, medical record/literature analysis, and disease prediction management.

According to the 2020 AI Medical Industry Development Blue Book by China Academy of Information and Communications Technology,Although the domestic AI medical field started late, the market demand is strong and the future development prospects are broad.Among them, it is worth noting that as of the end of 2019, the proportion of the elderly population aged 65 and above in the country has reached 12.6%, which means that China has officially entered an aging society. As a result, the incidence of chronic diseases is also increasing year by year.

In this context, disease prediction-related results represented by this study have emerged, which can effectively help doctors and patients better manage their health. However, on the other hand, it is also necessary to see that from the overall market situation, AI-related technologies have not yet been applied on a large scale in hospitals, and hospitals are not willing to pay. This is closely related to the user's usage and payment habits, supporting infrastructure such as medical insurance policies, and the high complexity of clinical application scenarios.Therefore, there is still a long way to go in the field of AI medical care.

Reference Links:

[1] https://doi.org/10.5334/gh.934

[2] https://doi.org/10.1111/1753-0407.13175

[3] https://doi.org/10.1007/s001250051352

[4] https://doi.org/10.1186/1475-2840-5-15

[5]https://rs.yiigle.com/CN112148202107/1328929.htm