Massachusetts General Hospital

Traditional preoperative risk tools like ACS NSQIP require manual data entry and have limited accuracy in some subgroups, while newer AI risk models frequently go unused because each individual clinician must learn, evaluate, and safely incorporate them independently. In a high-volume GI surgery department with 30+ surgeons, this distributed AI adoption model was fragile, inconsistent, and unsustainable.

The team deployed the FLEX Score — an AI model using routinely collected preoperative EHR data to predict postoperative mortality, 30-day readmission, length of stay, and discharge to non-home care — within a centralized workflow. A single trusted surgeon paired with one research assistant reviewed all flagged high-risk patients weekly, using FLEX Score outputs alongside focused chart reviews to refer the highest-risk cases to the hospital's Perioperative Optimization of Senior Health clinic for geriatric assessment and interdisciplinary optimization.

Over a 4-month period from May to August 2025, the centralized team screened more than 2,200 scheduled GI surgeries across a department of 30+ surgeons using only one surgeon and one research assistant. The FLEX Score reduced the weekly review burden from 100+ scheduled patients to approximately 10 highest-risk patients, enabling department-wide scale without proportional increases in personnel. Buy-in from surgeons was rapid — the workflow expanded from a 2-month colorectal pilot to the full GI department within days.

• Centralizing AI use to a small, clinically credible team overcomes adoption barriers more effectively than distributing AI tools to every individual clinician.
• Automated risk stratification can achieve department-level scale without proportional personnel growth — one surgeon and one assistant covered 30+ surgeons across 2,200+ cases.
• EHR vendor integration remains a critical bottleneck; without native EHR support, AI tools must operate externally, limiting scalability and sustainability.