Risk Stratification Models: How Well do They Predict Adverse Outcomes in a Large Dutch Bariatric Cohort?

Abstract

Background: Risk prediction models are useful tools for informing patients undergoing bariatric surgery about their risk for complications and correcting outcome reports. The aim of this study is to externally validate risk models assessing complications after laparoscopic Roux-en-Y gastric bypass (LRYGB) surgery.

Methods: All 740 patients who underwent a primary LRYGB between December 2007 and July 2012 were included in the validation cohort. PubMed was systematically searched for risk prediction models. Eight risk models were selected for validation. We classified our complications according to the Clavien-Dindo classification. Predefined criteria of a good model were a non-significant Hosmer and Lemeshow test, Nagelkerke R (2) ≥ 0.10, and c-statistic ≥0.7.

Results: There were 85 (7.8 %) grade 1, 54 (7.3 %) grade 2, 5 (0.7 %) grade 3a, 14 (1.9 %) grade 3b, and 14 (1.9 %) grade 4a complications in our validation cohort. Only one model predicted adverse events satisfactorily. This model consisted of one patient-related factor (age) and four surgeon- or center related factors (conversion to open surgery, intraoperative events, the need for additional procedures during LRYGB and the learning curve of the center).

Conclusions: The overall majority of the included risk models are unsuitable for risk prediction. Only one model with an emphasis on surgeon- and center-related factors instead of patient-related factors predicted adverse outcome correctly in our external validation cohort. These findings support the establishment of specialty centers and warn benchmark data institutions not to correct bariatric outcome data by any other patient-related factor than age.

Keywords: Adverse events; Bariatric surgery; Benchmark data; Complications; Gastric bypass; Morbid obesity; Outcome reporting; Risk factors; Risk stratification; Validation.