AORTA: A Model Predicting Ascending Aortic Diameter From Clinical Variables

A study published in JAMA formulated a predictive tool, AORTA (aorta optimized regression for thoracic aneurysm), to estimate the ascending aortic diameter on the basis of clinical characteristics.

Investigators from Massachusetts General Hospital sourced data for this study from the UK Biobank (n=36699), Framingham Heart Study (FHS; n=1367), and Mass General Brigham’s (MGB) Community Care Cohort Project (n=50,768). Demographic and clinical variables and data from magnetic resonance imaging (MRI), computed tomography, and transthoracic echocardiography were used in a least absolute shrinkage and selection operator (LASSO) regression model to develop the AORTA estimate of the ascending aortic diameter. Data from the Biobank were subdivided into derivation (n=30,018) and validation (n=6681) cohorts and the FHS and MGB cohorts were used as external validation. Missing data were imputed.

Among the study cohorts, the proportion of men was 42.6% to 48.4%, they had a median age of 59.0 to 65.1 years, they had BMIs of 25.7 to 28.0, 30.0% to 76.2% had hypertension, 16.1% to 21.9% had hyperlipidemia, and 2.9% to 27.0% had diabetes.

Among the derivation cohort, at MRI the mean ascending aortic diameter was 3.18 (SD, 0.35) cm and median diameter was 3.15 (IQR, 2.93-3.40) cm.

The LASSO model retained all variables (age, gender, BMI, heart rate, systolic and diastolic blood pressures, height, weight, diabetes, hypertension, hyperlipidemia, and interaction terms) as important predictors in the variable selection phase.

The measured and predicted mean aortic diameters were 3.18 and 3.17 cm for the Biobank, 3.48 and 3.13 cm for the FHS, and 3.17 and 3.12 cm for the MGB validation cohorts, respectively.

The model was able to explain 28.2% of the variance in ascending aortic diameter for the Biobank validation set, had a correlation with the real-world measurement of 0.53, and had a mean absolute percentage error of 7.4%.

For the external validation cohorts, the model explained 30.8% and 32.6% of the variance, had a correlation with real-world measurements of 0.56 and 0.57, and a mean absolute percentage error of 10.2% and 9.1% for the FHS and MGB cohorts, respectively.

Among the subset of patients in the derivation cohort who had an ascending aortic diameter of 4 cm or wider (2.4%), the model had an area under the receiver operating characteristic curve (AUROC) for predicting enlarged aorta of 0.770. In the FHS and MGB cohorts, the model had an AUROC for predicting an enlarged aorta between 0.766 and 0.813.

Compared with other models for predicting aortic diameter, the AORTA score performed significantly better than the age-plus-sex model (P <.001) or the model from Obel, et al (P <.001).

Using data from the derivation cohort, an AORTA score cutoff of 3.537 cm was found to best predict aortic diameter of 4 cm or more.

The AORTA tool is potentially limited as estimates are based on blood pool-based diameters measured at 1 imaging plane and do not describe the entire ascending aorta.

“Further research is needed to optimize the prediction model and to determine whether its use is associated with improved outcomes,” the study authors wrote.

Disclosure: Multiple authors declared affiliations with industry. Please refer to the original article for a full list of disclosures.