Additionele informatie
BACKGROUND When counseling patients with intraductal papillary mucinous neoplasms (IPMN), the risks of current or future high-grade dysplasia and malignancy are weighed against the risks associated with surgery. However, a reliable tool to determine the risks of surgery in this setting is not available. This study identified predictors for mortality and major morbidity in patients undergoing surgery for IPMN as well as a risk model.
METHODS International, retrospective study including patients undergoing pancreatic surgery for IPMN within the audits from North America, Germany, the Netherlands, and Sweden (GAPASURG, 2018–2020) with the primary aim to determine preoperative predictors for in-hospital/30-day mortality and major morbidity.
Study design
This retrospective, international study was reported according to the TRIPOD+AI guidelines.[15] The study was endorsed and approved by the scientific committee of the Global Audits on Pancreatic Surgery (GAPASURG) collaboration.
Study population
All patients who underwent partial pancreatectomy – either pancreatoduodenectomy (PD) or left pancreatectomy (LP) – for IPMN and were recorded in the GAPASURG registry (2018–2020) were included. Patients were excluded in case of non-formal non-oncological pancreatic resections (e.g., enucleation procedures) or if data on the primary study endpoint was missing
Endpoints
The primary endpoints were in-hospital/30-day mortality and major morbidity. Other endpoints included the individual rates of pancreatic surgery-specific complications (i.e., postoperative pancreatic fistula (POPF), delayed gastric emptying (DGE), and post-pancreatectomy hemorrhage (PPH)) and preoperative patient characteristics such as sex, age, BMI, functional health status (World Health Organization Performance Status [WHO PS] or Eastern Cooperative Oncology Group [ECOG]) and comorbidities.
Definitions
All postoperative complications were evaluated during a 30-day postoperative follow-up. In-hospital events were registered when hospital stay exceeded >30 days, except within the American College of Surgeons National Surgical Quality Improvement Program (NSQIP), in which only 30-day follow-up was registered. Major morbidity was assessed using the Clavien–Dindo classification, with complications of grade 3 or higher considered major.[17] For the evaluation of pancreatic surgery-specific complications, the definitions of the International Study Group for Pancreatic Surgery (ISGPS) were used, including only clinically relevant (grade B/C) POPF, DGE, and PPH. Other postoperative outcomes included the rate of readmissions within 30-days and radiological or surgical reinterventions. Comorbidity was defined as the presence of preoperative diabetes mellitus, chronic obstructive pulmonary disease (COPD), chronic heart failure, or renal dialysis. For functional health status, the Eastern Cooperative Oncology group performance status was recategorized to the corresponding WHO performance status (PS) categories to allow for matching: independent (WHO PS 0–1), partially dependent (WHO PS 2–3), and totally dependent (WHO PS 4). In addition, the American Society of Anesthesiologists performance status (ASA) was used.
Data collection
All study variables were exported from the GAPASURG database (2018–2020). The GAPASURG registry is a global collaboration and comprises the following four surgical audits: NSQIP, Deutsche Gesellschaft für Allgemein- und Viszeralchirurgie- Studien-, Dokumentations- und Qualitätszentrum (DGAV StudoQ|Pancreas), the Dutch Pancreatic Cancer Audit (DPCA), and the Swedish National Pancreatic and Periampullary Cancer Registry (SNPPC).Differences in variables across the four audits due to different metric systems were resolved by converting data or recategorization to facilitate data harmonization. The complete process of combining the four registries was described by Mackay et al. Supplement 1 provides a detailed overview of the variables included in each registry.
Statistical analyses
Categorical variables were reported as numbers with percentages and as mean with standard deviation (SD) for normal distributed data and as median with interquartile ranges (IQRs) for non-normally distributed data. Groups were compared using Pearson’s chi-squared test for categorical variables and the Kruskal–Wallis test for continuous variables as appropriate.
Multivariable logistic regression analyses were performed to identify independent preoperative predictors for in-hospital/30-day mortality and major morbidity and estimate adjusted odds ratios (ORs) and 95% confidence intervals (95% CIs). To facilitate implementation of the model, several variables were dichotomized. The ASA score and WHO performance status were dichotomized as follows: ASA 1–2 vs ASA 3–4, WHO PS 0–1 vs WHO PS 2–4.
Following the TRIPOD+AI recommendations for prediction models, model performance was assessed in terms of its discrimination (area under the ROC curve [AUC]), calibration, and clinical utility. Calibration refers to the agreement between the predicted and observed risks of in-hospital/30-day mortality and major morbidity, assessed using flexible calibration curves alongside the calibration intercept and slope (ideal values, respectively, 0 and 1). Lastly, decision curve analysis was used to evaluate whether a ‘model-guided treatment approach’ offered either equal or greater clinical benefit compared with a ‘treat all’ and ‘treat none’ approach.
Internal validation was carried out using bootstrapping with 5000 bootstrap resamples. Each bootstrap sample underwent all modeling steps (including backward selection with P<0.05) to comprehensively account for all model uncertainty and ‘researcher degrees-of-freedom’. Bootstrap-corrected performance measures were estimated using the approach of Efron and Gong.
Statistical analyses were performed using RStudio, version 4.5.2, with the CalibrationCurves (v3.0.0), dcurves (v0.5.1), and rms (v8.1–0) packages. All p-values were two-tailed, and significance level was set at P<0.05.
RESULTS Among 2041 patients undergoing partial pancreatectomy for IPMN, 54.5% (n=1113) underwent pancreatoduodenectomy (PD) and 45.5% (n=928) left pancreatectomy. The rate of in-hospital/30-day mortality was 1.7% and of major morbidity 24.1%. Independent predictors of mortality were undergoing PD (adjusted odds ratio, 6.78 [95% CI, 2.37 to 19.3; P<0.0001)[CL1] , age ³65 years (4.89 [1.48 to 16.0]; P=0.001), WHO performance status ≥2 or higher (9.01 [2.41 to 33.7]; P=0.008). The prediction model for mortality had an AUC of 0.74 (95% CI, 0.67 to 0.80) with robust calibration. The risk of mortality ranged from 0% with no factors present to 24% with all factors present. The prediction model for major morbidity (AUC, 0.64 [95% CI, 0.61 to 0.67]) found a risk ranging from 12% with no factors present to 39% with all factors present. Decision curve analysis indicated that the clinical utility of a model-based treatment approach was either equal or superior to a ‘surgery-for-all’ and ‘surgery-for-none’ approach.
CONCLUSION Patients with IPMN at high risk for surgery-related mortality and morbidity can be reliably identified through the developed model which may support decision-making, and is available through www.pancreascalculator.com. Large studies are needed to confirm the clinical added value of this model.