Thorac cardiovasc Surg 2018; 66(08): e1-e2
DOI: 10.1055/s-0038-1639336
Letters to the Editor
Georg Thieme Verlag KG Stuttgart · New York

Developing a Risk Prediction Model for Intensive Care Unit Mortality after Cardiac Surgery

Qing Liu
1  Department of Anesthesiology, Plastic Surgery Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, People' Republic of China
,
Fu-Shan Xue
2  Department of Anesthesiology, Beijing Friendship Hospital, Capital Medical University, Beijing, People' Republic of China
,
Gui-Zhen Yang
1  Department of Anesthesiology, Plastic Surgery Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, People' Republic of China
,
Ya-Yang Liu
1  Department of Anesthesiology, Plastic Surgery Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, People' Republic of China
› Author Affiliations
Funding All the authors have no financial support.
Further Information

Publication History

15 January 2018

05 February 2018

Publication Date:
01 April 2018 (online)

Validation of Three Postoperative Risk Prediction Models for Intensive Care Unit Mortality after Cardiac Surgery

Reply to “Developing a Risk Prediction Model for Intensive Care Unit Mortality after Cardiac Surgery”

The recent article by Howitt et al[1] validating three postoperative risk prediction models for intensive care unit (ICU) mortality after cardiac surgery was of great interest. They showed that discriminative ability of all three models for mortality on each postoperative day was good with areas under the receiver operating characteristic curve of > 0.8. However, calibration tests showed that the ratios of observed to expected mortality for all three studied models were < 0.65, indicating a bad overall agreement between predicted probabilities and observed frequencies in the development and the validation sets. Local recalibration had partly improved the predictive performance of three studied models, but significant under- and overestimation for daily postoperative mortality rate still existed, based on the results provided in their Table 4. This issue may be attributable to the following factors.

First, in the modern era, early mortality rate after cardiac surgery is very low. For example, in the Howitt et al's study including a total of 2,255 patients who underwent various cardiac surgery, early mortality rate in the ICU only was 1.8%.[1]

Second, the cause of early death after cardiac surgery is often multifactorial. Besides the risk factors included in the three studied models, other known factors have been also associated with the increased mortality rate after cardiac surgery. For example, in the cardiovascular system, postoperative atrial fibrillation occurs in ∼35% of cardiac surgery cases and has a peak incidence on postoperative day 2. The available evidence shows that postoperative atrial fibrillation is an independent predictor of numerous adverse postoperative outcomes, including a two- to fourfold increased risk of stroke, cardiac arrest, cerebral complications, need for permanent pacemaker placement, and a twofold increase in all-cause 30-day and 6-month mortality.[2] Furthermore, postoperative bleeding is common after cardiac surgery. Even among patients with stable hemodynamics, severe postoperative bleeding has also been associated with significantly worse outcomes after cardiac surgery.[3] In the blood system, the three studied models only include the postoperative platelet count to assess coagulation function. In fact, postoperative anemia is common, frequently persists for months after cardiac surgery, and is associated with an impaired outcome. When hemoglobin is considered as a continuous variable, every 1 mg/dL decrease in postoperative hemoglobin level is associated with a 22% increase in all-cause early mortality rate (hazard ratio 0.78, 95% confidence interval 0.60 to 0.99, p = 0.034).[4] Moreover, massive postoperative blood transfusion has been shown as an independent predictor of increased early mortality rate after cardiac surgery.[5] In the digest system, only bilirubin concentration was used for the assessment of hepatobiliary function in the three studied models. Perhaps, the widely accepted Child–Pugh score can provide a more comprehensive evaluation for early hepatobiliary function after cardiac surgery, especially for patients with hepatic dysfunction.[6] Recent work also determines that gastrointestinal complications after cardiac surgery occur at a rate of 4.17% and increase inpatient mortality threefold.[7] Finally, both bloodstream infection and sepsis are the important contributors to early death after cardiac surgery, particularly among the healthiest patients.[8] [9]

The low early mortality rate after cardiac surgery combined with the lack of above identified risk factors would have decreased the discriminative ability of the three studied models for early postoperative deaths in the ICU. Thus, we agree with the authors that further studies are still needed to optimizing the predictive models of early mortality rate after cardiac surgery. To obtain a clinically useful model with high sensitivity, specificity, and positive predictive value, moreover, we argue that the known risk factors associated with early mortality rate after cardiac surgery should be taken into the model as much as possible.

Authors' Contributions

Qing Liu: This author had carefully read the article of Howitt et al,[1] analyzed their methods and data, suggested comment points, and drafted this letter to the editor. Qing Liu had seen and approved the final manuscript.


Fu-Shan Xue: This author had carefully read the article of Howitt et al,[1] analyzed their methods and data, revised the comment points and this letter, and is the author responsible for this manuscript. Fu-Shan Xue had seen and approved the final manuscript.


Gui-Zhen Yang: This author had read the article of Howitt et al,[1] helped to analyze their methods and data, and revised the comment points. Gui-Zhen Yang had seen and approved the final manuscript.


Ya-Yang Liu: This author had read the article of Howitt et al,[1] helped to analyze their methods and data, and revised this letter. Ya-Yang Liu had seen and approved the final manuscript.