ACS is one of the main causes of death worldwide. Predicting the mortality risk of ACS patients is helpful for disease management, prolonging the survival time of patients, and improving the quality of life of patients. In our study, the mortality rate was 1.75%, which was lower than that of the Asia-Pacific region; the mortality rate of ACS patients in the Asia-Pacific region during hospitalization was about 5%. Multivariate logistic regression analysis found age, cTnI, CK, NT-proBNP and LDH were independently associated with increased risk of in-hospital mortality. Moreover, calcium channel blockers and HDL cholesterol were independently associated with decreased risk of in-hospital mortality. The study showed that the models developed using logistic regression, GBDT, random forest, and SVM algorithms can predict the risk of in-hospital mortality of ACS patients. We also found that NT-proBNP, LDH, and HDL cholesterol were top three variables that contribute the most to the prediction performance of GBDT model and random forest model.
We employed four machine learning algorithms to develop prediction models of mortality during hospitalization in ACS patients and compared their performances. The study found that prediction models constructed by logistic regression, GBDT, random forest, and SVM can effectively predict the risk of death for ACS patients in hospitals. Previous studies have found that the prediction performance of a 3-year mortality risk prediction model for ACS patients based on machine learning algorithms was better than the GRACE score (AUC: 0.768 vs. 0.701). Sherazi et al. found that a mortality risk prediction model based on machine learning algorithms can effectively predict the 1-year death risk of ACS patients. Another study found that a prediction model based on machine learning algorithms can predict the 30-day mortality rate of post-ST-segment elevation myocardial infarction. The authors suggested that machine learning can be used for outcome prediction in complex cardiology settings.
The risk of mortality of ACS patients during hospitalization is affected by risk factors. We evaluated the contribution of variables to the predictive effect of the GBDT and random forest risk prediction model.
The top three variables that contributed the most to the prediction effect of GBDT and random forest models were the levels of NT-proBNP, LDH, and HDL cholesterol. It was consistent with previous research that has found that among patients with chronic heart failure, patients with increased levels of NT-proBNP had a poorer prognosis. Furthermore, preoperative NT-proBNP can predict the in-hospital mortality and long-term survival of patients undergoing surgery for ACS. Another study found that in non-ST elevation-acute coronary syndrome (NSTE-ACS) patients, NT-proBNP had a good predictive effect on 30-day mortality (AUC = 0.85). Moreover, elevated plasma LDH was associated with worse outcomes and increased risk of mortality in patients with several diseases[28–32]. Previous studies have shown that increased plasma LDH level was an independent predictor of the risk of mortality in patients with acute aortic syndromes. Additionally, plasma LDH levels were associated with 28-day risk of death in patients with sepsis, and in patients with acute decompensated heart failure, plasma LDH may be an independent predictor of 90-day, 180-day, and 365-day all-cause mortality risk. However, there were few studies conducted specifically with ACS patients. Our study found that LDH was related to the increased risk of mortality in patients with ACS and that it provided a considerable contribution to the predictive effect of the model. Previous studies have found that low levels of HDL and LDL cholesterol have been shown to be important predictors of in-hospital mortality[36–38]. A low early HDL cholesterol level should be regarded as an independent predictor of in-hospital mortality in ACS patients presenting to the cardiac care unit. TRILOGY ACS Trial found that lower baseline HDL cholesterol was significantly associated with increased risk of cardiovascular death and all-cause death in ACS patients. The findings of both of these studies are consistent with the results of our study.
In addition to the above three important variables, age, cTnI, CK, D-dimer, LDL cholesterol, beta-blockers, and aspirin all had an important influence on the prediction effect of the model. These results were consistent with previous findings about these factors. Cardiac troponin I (cTnI) was a validated biomarker for diagnosis and risk stratification of patients with acute coronary syndrome. The increase of high sensitivity-cTnI level in the stable phase after ACS event was an independent predictor of all-cause death and cardiovascular death in the ACS outpatient population. In a stabilized phase of patients with non-ST-segment elevation ACS, cTnI levels exhibited a continuous and slight increase, and levels of cTnI elevated above 0.01 ug/L can predict the long-term mortality of patients. EPICOR registry study found that old age was a risk factor for poor prognosis during the first two years after discharge in ACS patients. As the age of ACS patients increases, the risk of poor prognosis increases, as even elderly patients with good heart function had a higher risk of death. The study found that compared with younger patients, elderly patients with ACS had a higher risk of comorbidities, hospitalization, and 6-month mortality. The increased D-dimer level was found to be a predictor of a patient's adverse outcome. Several studies have found that increased D-dimer levels were significantly associated with adverse outcomes and increased risk of mortality in ACS patients[47–49]. In addition, studies have found that increased levels of creatine kinase-MB can predict the risk of hospital death in elderly ACS patients. European Society Of Cardiology guidelines recommend that patients with NSTEMI and STEMI receive optimal medical therapy, which includes aspirin, beta-blockers, and other drugs. The study found that ACS patients receiving aspirin treatment before admission had a reduced 30-day mortality. Real-world studies have found that receiving optimal medical therapy (including aspirin and beta- blockers) in patients with ACS after discharge can reduce their mortality risk.
In our research, we found an interesting result. Calcium channel blockers was significantly associated with a reduction in the risk of hospital death in ACS patients. This was inconsistent with some research results in western countries. Several studies have found that calcium channel blockers had no advantages over other antihypertensive drugs in reducing the serious complications of hypertension, and calcium channel blockers increased the overall mortality risk and the adverse events risk in patients with coronary heart disease[54–56]. However, in the Japanese study, there was no significant difference in the incidence of cardiovascular death, reinfarction, uncontrolled unstable angina and nonfatal stroke in patients with post-acute myocardial infarction after receiving beta blockers and calcium antagonists. Another Japanese retrospective study found that Calcium channel blockers (nifedipine-retard) did not increase the incidence of cardiac events in post-MI patients, and even prevent the risk of cardiac events in non-smokers under 50 years of age. An Australian study found that calcium channel blockade was not associated with the excess risk of death in post-AMI patients.
Our research results suggested that preventing abnormally elevated levels of NT-proBNP, LDH, cTnI, and CK in patients with ACS, while preventing the level of HDL cholesterol in patients from falling below the normal range, may reduce the risk of in-hospital death in patients with ACS.
This study had several limitations. First, this was a retrospective study, and the results need to be verified by a prospective clinical trial. Second, our study was a single-center study; therefore, the results of the study limit the generalizability to apply to all ACS patients.