Risk of Heart Valve Surgery

Abstract & Commentary

By Michael H. Crawford, MD, Editor

Synopsis: Based on a large national database of heart valve patients, a risk score has been developed to predict operative mortality for any valve surgery.

Source: Ambler G, et al. Generic, Simple Risk Stratification Model for Heart Valve Surgery. Circulation. 2005;112: 224-231.

The incidence of heart valve surgery is rising, yet there is little specific information about the risk of valve surgery. Thus, Ambler and colleagues in the United Kingdom obtained data on over 32,000 patients from 30 institutions who underwent heart valve surgery between 1995 and 2003 in the United Kingdom and Ireland. Based upon the first 5 years, a data model was developed for predicting in-hospital risk of death (n = 16,679). The performance of the model was evaluated in the remaining 16,160 patients. The final model included data from both the development and testing groups, and is adjusted for the last year of the data, so that it will be as applicable as possible to current surgical practices. The total patient database had a 6.4% mortality, a mean age of 65 years, 42% were female, 64% had isolated aortic, and 29% isolated mitral valve surgery. Few patients had tricuspid (TV) or pulmonic valve surgery concomitantly, but about one-third had concomitant coronary artery bypass graft surgery (CABG).

Among the clinical characteristics evaluated for their predictive ability, only respiratory diseases were dropped. The area under the receiver operating curve of the model was 0.77, which suggests a reasonable ability to predict the outcome of this complex intervention. The risk predictors in order of importance were operative priority (urgent, emergency, etc.), age, renal failure (creatinine > 2.0), operation sequence (first, second, etc.), ejection fraction, concomitant TV surgery, type of valve operation (aortic, mitral, combined, etc.), concomitant CABG, body mass index, pre-operation arrhythmias (atrial fibrillation, ventricular tachycardia, etc.), diabetes, gender, and hypertension (all P < 0.01). Ambler et al then developed a risk score which can be used to estimate in-hospital mortality risk.

Commentary

As percutaneous interventions increase, the number of CABGs decreases, and valve surgery becomes a larger percent of heart surgeries—now about one-third. Also, emphasis on valve repair has increased the number of candidates for valve surgery. I usually quote patients the overall reported mortality for valve surgery of 4-8%, but mitral valve repair in a 50-year-old or aortic valve replacement in an 85-year-old, may have a risk below or above this range, respectively. In addition, when the surgeons are keen on operating, the risk is quoted as 1-3% and, when they are not, 25-50%. Until now, there has been no large, well-validated model that is easy to use for the cardiologist to come up with their own figure. Other strengths of this model are that it can be used for single valve cases, multiple valve cases, aortic or mitral, and with or without concomitant CABG. Because of the robust design of the model, developing separate models for stenosis and regurgitation or for mitral and aortic valves made no significant difference in the risk predictions.

The top 4 predictors out of the 13 are the strongest: operative priority, renal failure, age, and operative sequence. This agrees with previous studies. Also, respiratory diseases dropped out as a predictor, which agrees with prior data. One weakness of the model is that the presence of infective endocarditis is not included because there was insufficient data on this issue. Such cases only comprised about 4% of valve surgeries. The model has a point range of 0 to 25, with corresponding in-hospital mortality risks of 0.2 to 52.9%. This provides for the risk prediction scale commonly used by surgeons when talking to patients I discussed above (1-50%). The model was developed in the United Kingdom and Ireland, which have national health services, where it is believed the sickest patients languish, waiting weeks for surgery. So there may be a natural selection bias to their database. Finally, Ambler et al point out that this model is only warranted for 2003. New developments in surgery will have to be accounted for in later iterations. (The scoring sheet and risk calculation tables are clearly printed in the paper. Perhaps they will develop a PDA version.)