AI Helps Researchers Predict Heart Attack Mortality

Acute myocardial infarction (AMI) is becoming an increasingly common condition. It is expected that by 2035, roughly half of the adult population will suffer from some form of this disease.

Researchers have been examining various algorithms, curious about how accurately they can predict patient mortality/survival within one year. They worked with a dataset from MIT MIMIC-III – which is supposedly freely available, but you still need certain certifications and approval from an expert committee. In any case, the dataset contains data from 58,000 hospitalisations involving 40,000 patients (not just concerning AMI).

In the tests, two models performed the best – logistic model trees (LMT) and the Simple Logistic (SL) algorithm. Their prediction accuracy in this case was shown to be just over 85%.

Sources:
• Article: https://venturebeat.com/…/researchers-use-ai-to-predict-he…/
• Dataset: MIMIC-III https://www.nature.com/articles/sdata201635
• Request for dataset: https://mimic.physionet.org/gettingstarted/access/
Note: It is necessary to create an account on PhysioNetWorks and have a CITI certificate for "Data Only or Specimens Only Research" beforehand.