Article 1
Effect of surgical timing on outcomes after cholecystectomy for mild gallstone pancreatitis. Cho NY, Chervu NL, Sakowitz S, Verma A, Kronen E, Orellana M, de Virgilio C, Benharash P. Surgery. 2023 Jun 22:S0039-6060(23)00304-5.
Gallstone pancreatitis accounts for 270,000 ED visits, an estimated $2.2 billion in annual costs, and remains one of the most common diagnoses managed by Emergency General Surgeons. Despite previous single center studies advocating for early cholecystectomy in patients with mild gallstone pancreatitis, the correlation between delayed cholecystectomy and adverse events in these patients has remained in question. Additionally, the financial impact of delayed cholecystectomy and overutilization of pre-operative ERCP has not been widely studied. The primary purpose of this study by Cho et al. was to evaluate the risk of major adverse events in patients who underwent early cholecystectomy (within 2 days) vs. late cholecystectomy (more than 2 days) for mild gallstone pancreatitis, defined as pancreatitis without end-organ dysfunction. Additionally, the authors sought to examine the financial burden of late cholecystectomy.
Using the National Readmissions Database, over 129,000 patients from 2016-2019 with mild gallstone pancreatitis were identified. Multivariable regression analysis and cubic spline modeling demonstrated an inflection point at 2 days for major adverse events and 30-day readmissions. Only 26% of patients underwent cholecystectomy within 2 days of admission. Patients in the late cholecystectomy cohort were older, more commonly male, and had a higher mean Elixhauser Comorbidity Index Score compared to the early cholecystectomy patients. Patients undergoing late cholecystectomy had higher rates of preoperative ERCP compared to early cholecystectomy (22 vs. 11%, respectively). On multivariate analysis, late cholecystectomy was associated with increased risk of major adverse events (AOR 1.4), including cardiovascular, infectious, and respiratory complications. Also, late cholecystectomy was associated with higher hospitalization costs (AOR 2.53), non-home discharge (AOR 1.41), and 30-day non-elective readmissions (AOR 1.12). In subgroup analysis, the authors did identify a risk reduction for major adverse events and cost reduction of $3,300 in hospitals with high cholecystectomy volumes compared to low volume hospitals.
Despite several recent studies demonstrating the safety and efficacy of early cholecystectomy, this study highlights the fact that cholecystectomy for mild gallstone pancreatitis quite commonly occurs more than 48 hours after admission. The data suggest that cholecystectomy within 2 days of admission is associated with fewer major adverse events, 30-day readmissions, and reduces hospitalization costs. The authors comment on several limitation that must be considered, including the retrospective design of the study using a large administrative database. Granular data, including vital signs, laboratory values, and BMI, were unavailable, which limited the ability to estimate pancreatitis severity. Additionally, the indication for ERCP and several patient and/or surgeon specific factors could not be determined, which may influence the results of the study. Nevertheless, these results suggest the standard of care for mild gallstone pancreatitis is yet to be defined and that cholecystectomy within 2 days of admission should be prioritized by Emergency General Surgeons.
Article 2
Artificial Intelligence versus Surgeon Gestalt in Predicting Risk of Emergency General Surgery. El Moheb M, Gebran A, Maurer LR, Naar L, El Hechi M, Breen K, Dorken-Gallastegi A, Sinyard R, Bertsimas D, Velmahos G, Kaafarani HMA. J Trauma Acute Care Surg. 2023 Jun 14.
The prediction of surgical morbidity and mortality has drastic implications in clinical and family decision making as well as risk-adjustment for quality benchmarking. The determination of risk for the development of major complications following emergency general surgery is complex and does not follow a linear prediction model. The authors of this study leveraged Artificial Intelligence technology and had previously developed a non-linear risk calculator called Predictive OpTimal Trees in Emergency General Surgery Risk (POTTER) to predict post-operative outcomes in emergency surgery patients. The objective of this study was to compare POTTER to surgeons’ surgical risk estimation and to assess how POTTER influences surgeon’s risk estimation.
150 patients who underwent emergent laparotomy for a general surgery pathology (trauma, vascular, and gynecological laparotomies excluded) at an academic tertiary referral center were followed prospectively for 30-day post-operative complications (mortality, septic shock, ventilator dependence, bleeding requiring transfusion, and pneumonia) using standard NSQIP definitions for complications. Clinical cases were then created based on these 150 de-identified patients using a standard template for the assessment of complication prediction. 30 acute care surgeons with diverse experience and practice settings were then randomized to make predictions of complication rate for the clinical scenarios based solely on their judgement (SURG group) or with access to POTTER’s predictions (SURG-POTTER group) and the surgeons’ predictions were compared to POTTER’s predictions.
The SURG group had significantly higher mean predicted estimates for all outcomes (mortality, septic shock, ventilator dependence, bleeding, and pneumonia) compared to both POTTER and SURG-POTTER. With regard to predicting outcomes, POTTER outperformed SURG in mortality (AUC: 0.880 vs 0.841), ventilator dependence (AUC: 0.928 vs 0.833), bleeding (AUC: 0.832 vs 0.735) and pneumonia (AUC: 0.837 vs 0.753). SURG-POTTER outperformed SURG in predicting mortality (AUC: 0.870 vs 0.841), bleeding (AUC: 0.811 vs 0.735) and pneumonia (AUC: 0.803 vs 0.753). POTTER did not outperform SURG in predicting septic shock and SURG-POTTER did not outperform SURG in predicting septic shock or ventilator dependence.
This study demonstrates the enhancement AI brings to the inherent complexity of risk calculation by deviating from the traditional linear modeling most risk scores and calculators are based on. More importantly, it shows how using an AI tool to complement the surgeon’s bedside assessment of risk made the surgeon less likely to overestimate mortality and complications, more accurate in predicting outcomes, and provided less variability in risk estimates. AI tools like POTTER can improve the accuracy of risk prediction and allow for enhanced informed decision making.
