Predictive models for mortality after ruptured aortic aneurysm repair do not predict futility and are not useful for clinical decision making

• Published in: Journal of vascular surgery 2016-12, Vol.64 (6), p.1617-1622
• Main Authors: Thompson, Patrick C., MD, Dalman, Ronald L., MD, Harris, E. John, MD, Chandra, Venita, MD, Lee, Jason T., MD, Mell, Matthew W., MD, MS
• Format: Article
• Language: English
• Subjects: Aged; Aged, 80 and over; Algorithms; Aortic Aneurysm, Abdominal - diagnosis; Aortic Aneurysm, Abdominal - mortality; Aortic Aneurysm, Abdominal - surgery; Aortic Rupture - diagnosis; Aortic Rupture - mortality; Aortic Rupture - surgery; California; Chi-Square Distribution; Databases, Factual; Decision Support Techniques; Female; Hospitals, University; Humans; Male; Medical Futility; Middle Aged; Patient Selection; Predictive Value of Tests; Reproducibility of Results; Retrospective Studies; Risk Assessment; Risk Factors; Surgery; Time Factors; Treatment Outcome; Unnecessary Procedures; Vascular Surgical Procedures - adverse effects; Vascular Surgical Procedures - mortality
• ISSN: 0741-5214; 1097-6809
• DOI: 10.1016/j.jvs.2016.07.121

Abstract Objective The clinical decision-making utility of scoring algorithms for predicting mortality after ruptured abdominal aortic aneurysms (rAAAs) remains unknown. We sought to determine the clinical utility of the algorithms compared with our clinical decision making and outcomes for management of rAAA during a 10-year period. Methods Patients admitted with a diagnosis rAAA at a large university hospital were identified from 2005 to 2014. The Glasgow Aneurysm Score, Hardman Index, Vancouver Score, Edinburgh Ruptured Aneurysm Score, University of Washington Ruptured Aneurysm Score, Vascular Study Group of New England rAAA Risk Score, and the Artificial Neural Network Score were analyzed for accuracy in predicting mortality. Among patients quantified into the highest-risk group (predicted mortality >80%-85%), we compared the predicted with the actual outcome to determine how well these scores predicted futility. Results The cohort comprised 64 patients. Of those, 24 (38%) underwent open repair, 36 (56%) underwent endovascular repair, and 4 (6%) received only comfort care. Overall mortality was 30% (open repair, 26%; endovascular repair, 24%; no repair, 100%). As assessed by the scoring systems, 5% to 35% of patients were categorized as high-mortality risk. Intersystem agreement was poor, with κ values ranging from 0.06 to 0.79. Actual mortality was lower than the predicted mortality (50%-70% vs 78%-100%) for all scoring systems, with each scoring system overestimating mortality by 10% to 50%. Mortality rates for patients not designated into the high-risk cohort were dramatically lower, ranging from 7% to 29%. Futility, defined as 100% mortality, was predicted in five of 63 patients with the Hardman Index and in two of 63 of the University of Washington score. Of these, surgery was not offered to one of five and one of two patients, respectively. If one of these two models were used to withhold operative intervention, the mortality of these patients would have been 100%. The actual mortality for these patients was 60% and 50%, respectively. Conclusions Clinical algorithms for predicting mortality after rAAA were not useful for predicting futility. Most patients with rAAA were not classified in the highest-risk group by the clinical decision models. Among patients identified as highest risk, predicted mortality was overestimated compared with actual mortality. The data from this study support the limited value to surgeons of the currently published algo