Development and validation of an artificial neural network algorithm to predict mortality and admission to hospital for heart failure after myocardial infarction: a nationwide population-based study

Abstract

Background Patients have an estimated mortality of 15-20% within the first year following myocardial infarction and one in four patients who survive myocardial infarction will develop heart failure, severely reducing quality of life and increasing the risk of long-term mortality. We aimed to establish the accuracy of an artificial neural network (ANN) algorithm in predicting 1-year mortality and admission to hospital for heart failure after myocardial infarction. MethodsIn this nationwide population-based study, we used data for all patients admitted to hospital for myocardial infarction and discharged alive from a coronary care unit in Sweden (n=139 288) between Jan 1, 2008, and April 1, 2017, from the Swedish Web system for Enhancement and Development of Evidence-based care in Heart disease Evaluated According to Recommended Therapies (SWEDEHEART) nationwide registry; these patients were randomly divided into training (80%) and testing (20%) datasets. We developed an ANN using 21 variables (including age, sex, medical history, previous medications, in-hospital characteristics, and discharge medications) associated with the outcomes of interest with a back-propagation algorithm in the training dataset and tested it in the testing dataset. The ANN algorithm was then validated in patients with incident myocardial infarction enrolled in the Western Denmark Heart Registry (external validation cohort) between Jan 1, 2008, and Dec 31, 2016. The predictive ability of the model was evaluated using area under the receiver operating characteristic curve (AUROC) and Youden's index was established as a means of identifying an empirical dichotomous cutoff, allowing further evaluation of model performance. Findings 139 288 patients who were admitted to hospital for myocardial infarction in the SWEDEHEART registry were randomly divided into a training dataset of 111 558 (80%) patients and a testing dataset of 27 730 (20%) patients. 30 971 patients with myocardial infarction who were enrolled in the Western Denmark Heart Registry were included in the external validation cohort. A first event, either all-cause mortality or admission to hospital for heart failure 1 year after myocardial infarction, occurred in 32 308 (23•2%) patients in the testing and training cohorts only. For 1-year all-cause mortality, the ANN had an AUROC of 0•85 (95% CI 0•84-0•85) in the testing dataset and 0•84 (0•83-0•84) in the external validation cohort. The AUROC for admission to hospital for heart failure within 1 year was 0•82 (0•81-0•82) in the testing dataset and 0•78 (0•77-0•79) in the external validation dataset. With an empirical cutoff the ANN algorithm correctly classified 73•6% of patients with regard to all-cause mortality and 61•5% of patients with regard to admission to hospital for heart failure in the external validation cohort, ruling out adverse outcomes with 97•1-98•7% probability in the external validation cohort.Interpretation Identifying patients at a high risk of developing heart failure or death after myocardial infarction could ...

Applying Artificial Intelligence to Wearable Sensor Data to Diagnose and Predict Cardiovascular Disease: A Review

“…A neural network (NN) is one way to train a computer to mimic the human brain, which enables the computer system to achieve AI through deep learning (DL). In recent years, there has been a surge in research publications where AI has been used in medicine and healthcare [ 20 , 21 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 , 47 , 48 , 49 , 50 , 51 , 52 , 53 , 54 , 55 , 56 , 57 , 58 , 59 , 60 , 61 , 62 , 63 , 64 , 65 , 66 , 67 , 68 , 69 , 70 , 71 ,…”

An Artificial Intelligence Approach to Guiding the Management of Heart Failure Patients Using Predictive Models: A Systematic Review

“…Recently, artificial intelligence (AI) methods are successfully implemented in several medical fields e.g., in radiological images analysis or in prediction of suicide attempts [ 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 ]. The same applies to heart failure population.…”

Medication optimisation in hospitalised older people with polypharmacy and multimorbidity

“…Medication reconciliation effectively decreases admission and discharge order discrepancies, possibly reducing preventable medication harm [73,74]. Hence, the integration of medication reconciliation by pharmacy technicians at transitions of care has been implemented as a standard of care for several years in Dutch hospitals [75,76].…”

“…The distribution of patients among the four participating trial sites was 399 (48.3%), 132 (16.0%), 92 (11.1%) and 203 (24.6%) for Switzerland, Belgium, Ireland, and the Netherlands, respectively. The study population had a median age of 78 (IQR [74][75][76][77][78][79][80][81][82][83][84]; the median number of co-morbidities was 11 (IQR 8-17), and the median number of medications was 10 (IQR 7-13). 8.4% of the study patients were nursing home residents, and the Barthel Index of Activities of Daily Living score [18] was high (median 95; IQR 75-100) (Table 1).…”

“…Another study demonstrated that artificial intelligence could identify patients at high risk of heart failure or death after myocardial infarction [76]. Thus, the availability of large, real-time patient data combined with computer sciences is a promising strategy for future differentiation between older people at high or lower risk for drug-related harm, thereby increasing the PPV of existing screening tools based on medication use.…”