Postoperative hemodynamic deterioration among cardiac surgical patients can indicate or lead to adverse outcomes. Whereas prediction models for such events using electronic health records or physiologic waveform data are previously described, their combined value remains incompletely defined. The authors hypothesized that models incorporating electronic health record and processed waveform signal data (electrocardiogram lead II, pulse plethysmography, arterial catheter tracing) would yield improved performance versus either modality alone.


Intensive care unit data were reviewed after elective adult cardiac surgical procedures at an academic center between 2013 and 2020. Model features included electronic health record features and physiologic waveforms. Tensor decomposition was used for waveform feature reduction. Machine learning–based prediction models included a 2013 to 2017 training set and a 2017 to 2020 temporal holdout test set. The primary outcome was a postoperative deterioration event, defined as a composite of low cardiac index of less than 2.0 ml min˗1 m˗2, mean arterial pressure of less than 55 mmHg sustained for 120 min or longer, new or escalated inotrope/vasopressor infusion, epinephrine bolus of 1 mg or more, or intensive care unit mortality. Prediction models analyzed data 8 h before events.


Among 1,555 cases, 185 (12%) experienced 276 deterioration events, most commonly including low cardiac index (7.0% of patients), new inotrope (1.9%), and sustained hypotension (1.4%). The best performing model on the 2013 to 2017 training set yielded a C-statistic of 0.803 (95% CI, 0.799 to 0.807), although performance was substantially lower in the 2017 to 2020 test set (0.709, 0.705 to 0.712). Test set performance of the combined model was greater than corresponding models limited to solely electronic health record features (0.641; 95% CI, 0.637 to 0.646) or waveform features (0.697; 95% CI, 0.693 to 0.701).


Clinical deterioration prediction models combining electronic health record data and waveform data were superior to either modality alone, and performance of combined models was primarily driven by waveform data. Decreased performance of prediction models during temporal validation may be explained by data set shift, a core challenge of healthcare prediction modeling.

Editor’s Perspective
What We Already Know about This Topic
  • Hemodynamic deterioration after cardiac surgery can range from easily reversable to severe sustained events and may lead to clinically relevant adverse outcomes
  • Clinicians currently rely on close clinical observation and experiential judgment to anticipate and treat such events
  • Little is known regarding machine learning approaches to real-time prediction after cardiac surgery based on data available at the bedside in the electronic health record or features extracted from commonly used physiologic monitoring devices
  • The authors have previously developed advanced signal processing techniques for feature extraction from lead II of the electrocardiogram, the invasive arterial waveform, and peripheral plethysmography
What This Article Tells Us That Is New
  • In this single-center, retrospective cohort study, the authors studied machine learning–based prediction models for postoperative hemodynamic deterioration using discrete electronic health record and continuous physiologic waveform data, alone or in combination, for 1,555 patients after cardiac surgery
  • All patients had pulmonary artery catheters placed during surgery per institutional protocol, allowing the thermodilution-derived cardiac index to be included as a key component of the composite hemodynamic endpoint
  • The best performing model in the training data set (2013 to 2017) used both data sources (area under the curve, 0.803) but was primarily driven by waveform data, suggesting that a black box waveform approach alone may have clinical utility in this setting
  • However, validation of these approaches in a later data set (2017 to 2020) showed substantially decreased performance (area under the curve, 0.709), most likely consistent with the phenomena of data set shift