Cardiovascular disease is the leading cause of death worldwide. Acute myocardial infarction is the most serious type of coronary artery disease due to its high incidence, sudden onset, rapid progression, and high mortality rate. In recent years, the proportion of NSTEMI in myocardial infarction has been increasing year by year.twenty threeIt is the most common type of myocardial infarction, which can severely impair a patient's daily life and even become life-threatening.24,25which brings great challenges to social medicine. Currently, the mainstream treatment for myocardial infarction is PCI, which dredges blocked coronary arteries and improves ischemic symptoms. However, patients are usually likely to be re-admitted due to recurrent myocardial infarction or myocardial infarction complications after myocardial infarction. Therefore, it is very important to effectively identify high-risk groups and reduce the re-admission rate of such patients as much as possible.
In recent years, the rapid development of artificial intelligence technology has also produced brilliant results in various industries. In the field of medicine, machine learning, a branch of artificial intelligence science, has been fully applied in clinical practice due to its characteristics of efficiently and accurately processing high-dimensional interactive information in large-scale data sets. Using common features in large-scale data sets to identify complex, multidimensional and nonlinear relationships between clinical features and predict various outcomes has shown unique clinical applications in many medical fields, including oncology.8 Neurology26In the cardiovascular field, impressive results have also been achieved, such as D'Ascenzo F.'s Artificial Intelligence Risk Prediction after Acute Coronary Syndrome (PRAISE).27.
At present, there are few models to predict readmission of NSTEMI patients after PCI. Therefore, in this study, we screened risk factors for readmission of NSTEMI patients after PCI based on multiple ML methods and established a prediction model. After selecting 1576 patients diagnosed with NSTEMI and treated with PCI, and collecting and screening clinical data, 1363 patients were finally included and divided into training and validation sets in a 7:3 ratio according to chronological order. In the training and validation sets, patients were divided into readmission and non-readmission groups according to whether they were readmitted (Fig. 1). To ensure the representativeness of the variables, three complementary methods were used to screen the variables, which made it easier to identify characteristic variables closely related to the prediction of outcome events. After selection, seven factors influencing readmission of NSTEMI patients after PCI were found. These included discharge outcome (easy, not easy), admission mode (ambulatory, non-ambulatory), communication ability (good, poor), CRP, TC, HDL, and LDL. Six ML algorithms, namely LR, DT, RF, SVM, XGBoost, and AdaBoost, were used to develop models, and the predictive efficiency of each model was evaluated by AUC, accuracy, sensitivity, and specificity. After comparison, the predictive model constructed by the LR algorithm was 0.749, with a sensitivity of 62.5% and a specificity of 78.1%. The accuracy was 75.6%, proving that the model had good discrimination power. Therefore, the LR algorithm was finally selected, and a nomogram was established based on it to predict the readmission risk of NSTEMI patients after PCI. To further evaluate the accuracy of the model, a calibration curve was used. As shown in Figure 6, the performance curve of the model showed good consistency with the calibration curve, basically fluctuating around the ideal curve, and the accuracy of the predictive model was good. To evaluate the practicability of the clinical risk prediction model, the clinical decision curve and clinical impact curve were drawn and analyzed, and the curves showed that the predictive model had clinical practical value and could benefit patients.
Previous studies have shown that models built with machine learning techniques are superior to traditional risk scoring models. For example, a study by Kwon JM proposed that a model built with the machine learning algorithm deep learning outperformed the GRACE risk scoring model in predicting in-hospital mortality in AMI patients (AUC: 0.905 vs. 0.851, P< 0.001); Qiu H also proposed that the risk model of acute kidney injury in patients with acute ST-segment elevation myocardial infarction after PCI constructed based on LASSO regression was superior to the Mehran score 2 scoring model (AUC: 0.840 vs. 0.674; P< 0.001). Therefore, to further verify the performance of this model, we compare it with common clinical risk assessment models such as the adjusted GRACE score, KAMIR score, and ACEF score. In comparison, the predictive ability of the Normograph model is better than the adjusted GRACE score (AUC: 0.767 vs. 0.645, Z = 4.486, P< 0.001), KAMIR score (AUC: 0.767 vs 0.678, Z = 3.321, P< 0.001) and ACEF score (AUC: 0.767 vs 0.591, Z = 6.341, P< 0.001) (Table 4). Therefore, the model we established has good predictive ability.
In this study, we comprehensively included indicators related to patient hospitalization. These indicators included basic hospitalization information such as admission route (emergency or outpatient), admission mode (ambulatory or nonambulatory), consciousness at admission (awake or nonawake), communication ability at admission (good or poor), and sleep status (normal or abnormal). This level of detail is very rare in previous studies. We believe that detailed collection of patient data is helpful for comprehensive evaluation of patients. In our study, admission mode (ambulatory or nonambulatory) and communication ability at admission (good or poor) were independent risk factors for readmission in patients with NSTEMI after PCI. This is consistent with our traditional recognition, and nonambulatory admission and poor communication skills at admission indicate poor overall condition of patients, indicating that cardiac function of patients with NSTEMI after PCI remains poor after discharge, which may affect the patient's readmission rate.
Regarding patient discharge status, we also focused on discharge outcome (palliative or nonpalliative).For typical low-risk patients, early discharge after PCI is a safe strategy.28patients are required to be discharged promptly once their condition stabilizes, i.e., discharged after improvement. However, in our study, we found that the proportion of discharge outcomes that did not improve in the readmission group was significantly higher than that in the non-readmission group. This suggests that discharge outcomes are a distinctive variable that affects readmission, which may be due to the fact that such patients do not fully recover during hospitalization and often have vascular, bleeding, or cardiac complications before discharge.29Early discharge increases the likelihood of readmission because more medication and longer hospital stays are required. Therefore, rather than discharging patients without improvement, providing consistent treatment during hospitalization plays a role in preventing readmission.30,31
Regarding patients' inflammatory indicators, we focused on CRP because studies have shown that inflammatory factors have certain value in predicting the prognosis of acute myocardial infarction.32The risk of adverse events after PCI is positively correlated with CRP levels.33In this study, CRP levels in the readmission group were significantly higher than in the non-readmission group, confirming the findings of the study by Carrero JJ et al.32CRP belongs to acute phase proteins and is mainly synthesized in the liver. When infection or tissue damage occurs, the synthesis in the liver increases significantly, and the CRP level also increases significantly. The more significant the structural changes in the heart, the worse the cardiac function of the patient, indicating that CRP is related to the occurrence of ventricular remodeling.34Therefore, for patients with NSTEMI after PCI, close attention should be paid to CRP indices to reduce readmission rates.
In this study, TC, LDL, and HDL were all considered risk factors for readmission in patients with NSTEMI after PCI, and the readmission group had higher TC, higher LDL, and lower HDL than the non-readmission group. This is consistent with our usual understanding that dyslipidemia is at the core of the development of coronary heart disease. The 2019 European Society of Cardiology (ESC)/European Atherosclerosis Society (EAS) lipid guidelines state that:35 2018 Chinese guidelines for the diagnosis and treatment of stable coronary artery disease36 We emphasize that the risk of progression of coronary artery disease is positively correlated with blood lipids, especially low-density lipoprotein (LDL). Related lipid indicators such as total cholesterol (TC), triglycerides (TG), low-density lipoprotein (LDL), and high-density lipoprotein (HDL) have always been considered important tools for the prevention and treatment of coronary artery disease in clinical practice, but high TC, high LDL, and low HDL are independent risk factors for cardiovascular disease.37TC measurement is a useful screening index for detecting high-risk groups for coronary artery disease. High cholesterol is a risk factor for coronary artery disease and is also the main substance that promotes atherosclerosis, which has an obvious impact on patient re-admission. HDL carries cholesterol to surrounding tissues where it can be converted into bile acids or directly excreted from the intestine.38This process was found to lower cholesterol levels and reduce the risk of re-hospitalization, further supporting the findings of Lin T et al.39This study further confirmed that HDL is a protective factor in atherosclerosis, and its level reflects coronary artery lesions and affects the rehospitalization rate. The effect of LDL is opposite to that of HDL. In the early stage of atherosclerosis, endothelial function is impaired, and foam cells are formed by phagocytosis of oxidized LDL by monocytes/macrophages, which is the central link in atherosclerosis.40Observational studies and randomized controlled trials have proven that LDL is one of the important factors affecting the prognosis and rehospitalization of coronary artery disease.41,42Therefore, lowering LDL has clear benefits in reducing readmission rates.
This study has the following advantages. First, the current study focuses on the analysis of NSTEMI, the main type of rare myocardial infarction. In addition, we select representative variables by different methods, establish prediction models, and compare the models. Second, this study includes many variables and incorporates various indicators of patient hospitalization, including both admission and discharge data. This comprehensive approach allows us to identify previously unknown factors that affect the readmission rate. Third, a concise and clear bar graph was created, and the related indicators are easy to obtain and have high clinical practicality, which helps to use limited medical resources for patients at high risk of readmission. And this study has important significance in reducing the readmission rate, medical burden, and social and economic burden on patients.
