Can Comprehensive Imaging Analysis with Analytic Morphomics and Geriatric Assessment Predict Serious Complications in Patients Undergoing Pancreatic Surgery?

Abstract

We aimed to determine whether comprehensive imaging analysis with analytic morphomics (AM) enhances or replaces geriatric assessment (GA) in risk-stratifying pancreatic surgery patients. One hundred thirty-four pancreatic surgery patients were identified from a prospective cohort. Sixty-three patients in the cohort had preoperative CT scans in addition to comprehensive geriatric assessments. CT scans were processed using AM. Associations with National Surgical Quality Improvement Program (NSQIP) serious complications were evaluated using univariate analysis and robust elastic net modeling to obtain AUROC curves by adding AM and GA measures to our previously defined clinical base risk model (age, body mass index, American Society of Anesthesiologists classification, and Charlson comorbidity index). NSQIP serious complications were associated with low psoas Hounsfield units (HUs) (p = 0.002), low-density (0 to 30 HU) psoas area (p = 0.01), visceral fat HU (p ≤ 0.001), visceral fat area (p = 0.009), subcutaneous fat HU (p = 0.023), and total body area (p = 0.012) on univariate analysis. Elastic net models incorporating the base model with geriatric assessment and psoas HU (AUC = 0.751), and AM alone (AUC = 0.739) have greater predictive value than the base model alone (AUC = 0.601). The model utilizing AM and GA in combination had the highest predictive value (AUC = 0.841). When combined, AM and GA improve prediction of NSQIP serious complications compared to either technique alone. The additive nature of these two modalities suggests they likely capture unique aspects of a patient’s fitness for surgery.