Update of a Model to Predict Outcomes after Endovascular Aneurysm Repair.

BACKGROUND

Risk assessment models must be continuously validated and updated to ensure that predictions remain valid. Here, the Endovascular Aneurysm Repair Risk Assessment Model, developed in 2008, is updated and improved.

METHODS

We used prospectively collected data from Australian patients who underwent elective endovascular aneurysm repair between 2009 and 2013 (n = 695). Data were provided by treating surgeons and the National Death Index. Key outcomes were early and midterm survival, early complications (endoleak, operative, and graft-related) and late complications (endoleak and graft-related). Multinomial logistic regression determined which preoperative variables best predicted each outcome. Area under Receiver Operating Characteristic curve (AUROC), model P-value and internal validation statistics were used to select the best model.

RESULTS

Ten preoperative variables were included in the modeling for 10 key outcomes. The most valid outcomes with AUROC>0.7 were 1- and 3-year survival, 30 and 90-day mortality, early and late endoleak (types I, III and IV) and type II endoleak (with an increase in sac size ≥5 mm). The 10 preoperative variables that contributed to outcome models were self-reported fitness, American Society of Anesthesiologists physical status score, history of stroke/transient ischemic attack, age, aneurysm angle, infrarenal neck length, white cell count, respiratory assessment, diabetes and statin therapy. Fitness alone statistically significantly predicted 30 and 90-day deaths better than any other preoperative variable; achieving high AUROCs (0.78 and 0.80), and high odds ratios (12.8 [95% CI: 1.5-110.4] and 18.1 [95% CI: 2.2-149]).

CONCLUSIONS

An updated interactive predictive model of outcomes after endovascular aneurysm repair has been created. Many of the variables used in the 2008 model continued to be significant, however, new variables including fitness and respiratory assessment, improved the model. The new model uses variables routinely collected preoperatively, and hence can better support surgeon-patient discussions prior to operation. Informing patients of potential risks or likely outcomes following elective surgery can assist with preoperative shared decision-making.