The study developed deep learning methods (recurrent neural networks) to predict several severe complications (mortality, renal failure with a need for renal replacement therapy, and postoperative bleeding leading to operative revision) in post cardiosurgical care in real time. 11,492 adult patients who underwent major open heart surgery at a tertiary care centre for cardiovascular diseases formed the main derivation dataset. The accuracy and timeliness of the deep learning model's forecasts was compared to the predictive quality of established standard-of-care clinical reference tools. Results were externally retrospectively validated with 5,898 cases from the MIMIC-III critical care database.
The predictions significantly outperformed the standard clinical reference tools, improving the absolute complication prediction AUC by 0.29 (95% CI 0.23–0.35) for bleeding, by 0.24 (0.19–0.29) for mortality, and by 0.24 (0.13–0.35) for renal failure (p < 0.0001 for all three analyses). The deep learning methods showed accurate predictions immediately after patient admission to the intensive care unit while clinical reference tools required several hours of post-surgical patient observation before reaching their peak performance.
These findings are noteworthy in that they use routinely collected clinical data exclusively, without the need for any manual processing. This property is especially encouraging for prospective deployment in critical care settings where a system could direct the staff's attention towards patients who are most at risk of experiencing complications.