Comprehensive Summary
This study looked at whether multimodal deep learning models could better predict serious heart (cardiovascular) and brain (cerebrovascular) complications within 30 days after non-cardiac surgery than standard models. Using data from over 165,000 patients, a multimodal gradient boosting machine (GBM) model analyzed ECG waveforms along with patient demographic details like age, sex, and surgery type. This study found that the model was more accurate than standard, baseline GBM models. Overall, this study concluded that multimodal deep learning models can improve risk prediction while using minimal patient information.
Outcomes and Implications
The medical community can utilize multimodal deep learning models that could better predict serious heart and brain complications after non-cardiac surgery, decreasing the likelihood of morbidity, mortality, and healthcare costs that can arise due to unexpected complications. In addition, these models require minimal patient information, making them more convenient than standard models.