Descriptive characteristics of the 61 patients are shown in . Preoperatively, 18% had nonalcoholic fatty liver disease, 30% had hypertension, 33% had depression, 63% had sleep apnea, 11% had type 2 diabetes mellitus, and 28% of the female patients had polycystic ovary syndrome. Linear growth was minimal in our patients. The mean ± SD change in height was 0.2 ± 0.8 cm (range −2.7 to 2.5 cm). No patients needed treatment for bone-related problems during the study period.
Descriptive Characteristics of Study Participants at the Time of Surgery
Patients had an average of 3 DXA scans (range: 2–6). The first DXA scan was obtained at the following times: before surgery (n = 22); within 3 months after surgery (n = 19); and between 3 and 6 months after surgery (n = 20). The time of the last scan varied (mean: 13.9 months [range: 2.8–26.8]).
Weight, WB BMC, and BMD z score decreased significantly over time after surgery (P < .0001 for all) (). Regression models were for weight (), gender, time, and time2; for BMC (), height, gender, time, and time2; and for z score (), age and time. Weight was not included in the models for BMC and z score so that we could evaluate the change in these variables over time separate from change in weight.
Weight (A), BMC (B), and BMD z score (C) over time after bariatric surgery. Lines represent the 61 patients. The bold line is the predicted value for the regression characterizing change over time.
Weight showed a pronounced curvilinear decline over time (time squared term, P < .0001) (). Weight loss was greatest in the first year and then stabilized. Some patients had slight increases in weight after 15 months. On the basis of the regression model, the predicted weight loss at 6, 12, 18, and 24 months after surgery was 41.4 ± 1.3, 58.4 ± 1.4, 61 ± 1.5, and 49.2 ± 2.1 kg, respectively. Gender was a significant covariate in the regression model with male patients losing more weight (6.4 kg) than female patients.
BMC showed a slight curvilinear decline over time (time-squared term, P < .0001) (). Predicted values on the basis of regression modeling for WB BMC were 2692 g before surgery, decreasing to 2552 g at 1 year after surgery (−5.2%) and 2494 g at 2 years after surgery (−7.4%). Height and gender were significant covariates in the regression model. Height was positively associated with BMC (18.95 g/cm; P < .0001), and male patients had more BMC than female patients (323.4 g; P = .0004). For the patients who may still be growing, including height in the regression models accounted for the differences in bone mass and density because of linear growth.
BMD z score decreased linearly over time (P < .0001) (). The predicted values for BMD z score on the basis of regression modeling were 1.5 before surgery, decreasing to 0.8 at 1 year after surgery and 0.1 at 2 years after surgery. Although the z score decreased over time, the z score remained above average for age throughout the 2-year follow-up period. Age was included in the regression model because it was negatively associated with BMD z score (−0.056 per year; P = .03).
Results of the regression models for weight, WB BMC, and BMD z score were similar when restricting the analyses to the subset of individuals (N = 34) that had at least 1 additional measurement >12 months after surgery.
We examined the potential effect of weight loss for mediating or “explaining” the skeletal changes after surgery. Because of the pronounced curvilinear trend in weight loss over time, we restricted this set of analyses to those measurements that occurred within the first 12 months after surgery. The correlation between change in weight and change in BMC was r = 0.31 (P = .02), and the correlation between change in weight and change in BMD z score was r = 0.05 (P = .73). The correlations with BMC and BMD z score were not different when expressing weight change in absolute terms or as a relative value (eg, percentage of total weight).
We then fitted weight in the regression models to predict BMD z score and BMC. When both weight and time after surgery were included in the regression model that predicted BMD z score, the regression coefficient for time after surgery was attenuated by 44% (−0.052 vs −0.029) and was no longer statistically significant (P = .20). Inclusion of weight in the model for BMC resulted in a 14% reduction in the regression coefficient for time (−11.15 vs 9.64); however, it remained statistically significant (P = .01).