The Society of Thoracic Surgeons 2018 Adult Cardiac Surgery Risk Models: Part 2-Statistical Methods and Results.

Document Type


Publication Date


Publication Title

The Annals of thoracic surgery




BACKGROUND: The Society of Thoracic Surgeons (STS) uses statistical models to create risk-adjusted performance metrics for Adult Cardiac Surgery Database (ACSD) participants. Because of temporal changes in patient characteristics and outcomes, evolution of surgical practice, and additional risk factors available in recent ACSD versions, completely new risk models have been developed.

METHODS: Using July 2011 to June 2014 ACSD data, risk models were developed for operative mortality, stroke, renal failure, prolonged ventilation, mediastinitis/deep sternal wound infection, reoperation, major morbidity or mortality composite, prolonged postoperative length of stay, and short postoperative length of stay among patients who underwent isolated coronary artery bypass grafting surgery (n = 439,092), aortic or mitral valve surgery (n = 150,150), or combined valve plus coronary artery bypass grafting surgery (n = 81,588). Separate models were developed for each procedure and endpoint except mediastinitis/deep sternal wound infection, which was analyzed in a combined model because of its infrequency. A surgeon panel selected predictors by assessing model performance and clinical face validity of full and progressively more parsimonious models. The ACSD data (July 2014 to December 2016) were used to assess model calibration and to compare discrimination with previous STS risk models.

RESULTS: Calibration in the validation sample was excellent for all models except mediastinitis/deep sternal wound infection, which slightly underestimated risk and will be recalibrated in feedback reports. The c-indices of new models exceeded those of the last published STS models for all populations and endpoints except stroke in valve patients.

CONCLUSIONS: New STS ACSD risk models have generally excellent calibration and discrimination and are well suited for risk adjustment of STS performance metrics.

Clinical Institute

Cardiovascular (Heart)






Center for Cardiovascular Analytics, Research + Data Science (CARDS)