Current eligibility criteria for bariatric surgery use arbitrarily chosen body mass index (BMI) (calculated as weight in kilograms divided by height in meters squared) thresholds, an approach that has been criticized as arbitrary and lacking evidence.
To verify the importance of BMI as a mortality predictor, to identify other important mortality predictors, and to construct a mortality prediction rule in a population eligible for bariatric surgery.
We studied individuals from a population-representative register who met contemporary eligibility criteria for bariatric surgery (BMI, ≥35.0 alone or 30.0-34.9 with an obesity-related comorbidity) from January 1, 1988, through December 31, 1998. We used binary logistic regression to construct a parsimonious model and a clinical prediction rule for 10-year all-cause mortality.
The United Kingdom General Practice Research Database, a population-representative primary care registry.
Fifteen thousand three hundred ninety-four patients aged 18 to 65 years.
Main Outcome and Measure
Ten-year all-cause mortality.
Mean (SD) age was 46.9 (11.9) years, BMI was 36.2 (5.5), and 63.2% of the patients were women. All-cause mortality was 2.1%, and mean follow-up duration was 9.9 years. The final model, which included age (odds ratio, 1.09 per year [95% CI, 1.07-1.10]), type 2 diabetes mellitus (2.25 [1.76-2.87]), current smoking (1.62 [1.28-2.06]), and male sex (1.50 [1.20-1.87]), had a C statistic of 0.768. Although BMI significantly predicted mortality (odds ratio, 1.03 per unit [95% CI, 1.01-1.05]), it did not improve model discrimination or calibration. We divided clinical prediction rule scoring into 4 tiers. All-cause mortality was 0.2% in tier 1, 0.9% in tier 2, 2.0% in tier 3, and 5.2% in tier 4.
Conclusions and Relevance
All-cause 10-year mortality in obese individuals eligible for bariatric surgery can be estimated using a simple 4-variable prediction rule based on age, sex, smoking, and diabetes mellitus. Body mass index was not an important mortality predictor. Further work is needed to define low, moderate, and high absolute risk thresholds and to provide external validation.