Development of bariatric surgery-specific risk assessment tool.

BACKGROUND: Administrative databases have increasingly been used to assess bariatric surgery outcomes, resulting in policy recommendations about bariatric practice. However, surgical outcomes must be risk adjusted to compare patients of varying potential risk fairly with those to whom the policies will apply. To date, the risk adjustment tools used for database analysis of bariatric surgical outcomes have been those designed for other purposes, and their sensitivity for bariatric outcomes has not been established. METHODS: Bariatric surgical procedures contained in the National Hospital Discharge Summary for 1993-2003 were assembled into a database. The standard set of Elixhauser co-morbidity variables used by the Agency for Healthcare Research and Quality were entered into the database. Those variables that were significantly associated with adverse outcomes were entered into a stepwise-elimination logistic regression equation, yielding a set of variables related to adverse outcomes from bariatric surgery. These were then prospectively applied to another database (the National Inpatient Survey) to determine their sensitivity for predicting outcomes and were compared with the commonly used Charlson score. RESULTS: The variables significantly correlating with bariatric adverse events included chronic pulmonary disease, hypertension, diabetes with chronic complications, fluid and electrolyte disorders, deficiency anemias, and depression. Age and male gender were also signficantly related to adverse events. The c-index (a correlative index, with .5 showing no, and 1, a perfect, relationship) for bariatric surgery mortality with the Charlson index is .52. For the Elixhauser-based system we developed, it is .72. CONCLUSIONS: We have developed a new risk-adjustment tool for bariatric surgery outcomes studies that use administrative databases. Its performance was clearly better than that of the commonly used Charlson co-morbidity score. Bariatric studies that have used the Charlson index should not be considered adequately risk adjusted.