• Julia R Berian, Lynn Zhou, Marcia M Russell, Melissa A Hornor, Mark E Cohen, Emily Finlayson, Clifford Y Ko, Ronnie A Rosenthal, and Thomas N Robinson.
• American College of Surgeons, Division of Research and Optimal Patient Care, Chicago, IL.
• Ann. Surg. 2018 Jul 1; 268 (1): 93-99.

ObjectiveTo explore hospital-level variation in postoperative delirium using a multi-institutional data source.BackgroundPostoperative delirium is closely related to serious morbidity, disability, and death in older adults. Yet, surgeons and hospitals rarely measure delirium rates, which limits quality improvement efforts.MethodsThe American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP) Geriatric Surgery Pilot (2014 to 2015) collects geriatric-specific variables, including postoperative delirium using a standardized definition. Hierarchical logistic regression models, adjusted for case mix [Current Procedural Terminology (CPT) code] and patient risk factors, yielded risk-adjusted and smoothed odds ratios (ORs) for hospital performance. Model performance was assessed with Hosmer-Lemeshow (HL) statistic and c-statistics, and compared across surgical specialties.ResultsTwenty thousand two hundred twelve older adults (≥65 years) underwent inpatient operations at 30 hospitals. Postoperative delirium occurred in 2427 patients (12.0%) with variation across specialties, from 4.7% in gynecology to 13.7% in cardiothoracic surgery. Hierarchical modeling with 20 risk factors (HL = 9.423, P = 0.31; c-statistic 0.86) identified 13 hospitals as statistical outliers (5 good, 8 poor performers). Per hospital, the median risk-adjusted delirium rate was 10.4% (range 3.2% to 27.5%). Operation-specific risk and preoperative cognitive impairment (OR 2.9, 95% confidence interval 2.5-3.5) were the strongest predictors. The model performed well across surgical specialties (orthopedic, general surgery, and vascular surgery).ConclusionRates of postoperative delirium varied 8.5-fold across hospitals, and can feasibly be measured in surgical quality datasets. The model performed well with 10 to 12 variables and demonstrated applicability across surgical specialties. Such efforts are critical to better tailor quality improvement to older surgical patients.

29 keywords...

hide…