Predicting complications following bariatric surgery: the diagnostic accuracy of available tools.

Abstract

Bariatric surgery is effective and safe obesity treatment. Obesity-related co-morbidities contribute to higher postoperative morbidity. Assessment of possible outcomes seems to be crucial in the qualification process and perioperative care. Recently, various tools predicting complications after bariatric surgery have been proposed. The objective of the study was to validate the performance of available diagnostic algorithms as the predictors of 30-day complications after bariatric procedures. Department of General Surgery, University Hospital, Poland. The literature review was done to identify available risk prediction models. The analysis included patients after Roux-en-Y gastric bypass or sleeve gastrectomy (SG). The probability of postoperative complications was calculated for each patient. The assessed endpoints were general and severe morbidity stratified with Clavien-Dindo classification. The relationship between predicted and observed outcomes was assessed by logistic regression. Diagnostic accuracy was evaluated by the area under the receiver operating characteristic curves and Hosmer-Lemeshow test. Out of 1329 patients, 65.31% were women with a mean body mass index of 45 kg/m2. The majority of patients underwent SG (75.02%). The most common co-morbidities were as follows: dyslipidemia, hypertension, and diabetes. General morbidity reached 8.43%, whereas severe morbidity was 2.78%. We identified 10 eligible models. Only Gupta and Metabolic and Bariatric Surgery Accreditation and Quality Improvement Program (MBSAQIP) calculators presented a significant association with any and severe complications and reached acceptable accuracy in predicting severe complications. None of the models achieved sufficient discrimination in predicting general morbidity. All models remained well-fitted. MBSAQIP and Gupta's calculators seem to be helpful in the assessment of severe complications after bariatric surgery. Further studies should focus on improving the predictive accuracy of existing models.