The overall reduction in mortality and the resolution of chronic diseases such as type 2 diabetes are substantial following bariatric surgery 
. However, the potential for serious complications is a barrier for patients and payers to utilizing the long-term advantages offered by bariatric surgery. The penalizing of hospitals for early readmissions is already underway, and several states are imposing mandates that call for further reductions in readmissions. Primary care providers and surgeons alike will see substantial decreases in reimbursements for readmitted patients, and it is therefore imperative that systems be in place to prevent the occurrence of readmissions. In this study, we have identified factors predictive of severe events requiring hospital readmission within 30 days of RYGB or AGB in the largest prospective bariatric cohort to date and have established that short-term risk for readmission is low for both procedures and risk profiles are largely unique to each procedure.
We observed a RYGB readmission rate nearly five times higher than AGB. Previously reported hospital readmission rates for bariatric surgery vary widely in the literature 
. Possible explanations for these disparities may be due to differences in the patient populations, the definition of a hospital readmission, the proportion of patients within a sample undergoing open versus laparoscopic procedures, or surgeon experience.
The higher readmission and mortality rates for RYGB relative to AGB might suggest that AGB is preferable, however such risks must be weighed with the treatment outcomes. RYGB has been shown to result in greater weight loss and superior improvement in comorbid illness 
. Our group recently reported on greater one-year improvements among patients with type 2 diabetes with respect to weight loss, hemoglobin A1
C, medication scores, and rates of diabetes resolution for RYGB patients compared to matched AGB controls 
. Short-term complications must be weighed against the long-term benefits and complications of each procedure.
The clearest predictors of readmission following RYGB were the use of the open surgical approach and prolonged length of stay. While open procedures are justified for certain complex cases 
, our results, when considered with previous research 
, suggest that laparoscopic techniques should be preferred to open surgery in the absence of contraindications for laparoscopy. Patients with previous histories of bariatric surgery or other anatomical abnormalities may be best suited for open surgeries and would understandably be at higher risk for readmission. However, 7.6% of hospitals (19 of 249 BSCOE) conducted >80% of their RYGB procedures using the open approach and accounted for 46.6% of all open procedures in the database, suggesting that the open approach may be often dictated by surgeon preference in these hospitals rather than case difficulty.
The influence of serious comorbid disease on readmission risk for RYGB patients is expected, though the causal pathway of elevated risk for African-Americans is less clear. We suspect that the association with race may have been confounded by unmeasured variables such as surgical preparation, social support, economic status, or dietary intake. For AGB, readmission risk factors were quite different from those identified with RYGB with the exception of prolonged length of stay and severe ASA score: disability status, asthma, male gender, history of DVT/PE, and the presence of OSA or GERD.
Surprisingly, the profiles of risk factors for readmission were almost entirely distinct for the AGB and RYGB procedures. Prolonged length of stay following surgery was one of the only factors that significantly predicted readmissions in both surgical populations in multivariate analysis. That procedure-specific risk factors contrast so greatly between the two procedures is an important finding potentially overlooked by prior investigations. Previous studies have chosen to pool patients across procedures for analysis assuming that the underlying risk factors were the same 
. While several of our results are complementary, the choice to aggregate surgical patients may account for some important differences in results. Some previous studies examining readmission rates have identified high BMI as a risk factor for readmission 
, while our own did not. Risk analyses that pooled patients from multiple procedures may have observed an artificial inflation of risk for high-BMI patients who tend to undergo RYGB, which has a significantly higher readmission rate than AGB. The relationship between BMI and readmission risk may also have been confounded by a less complete comorbidity profile in risk models, since many conditions are more prevalent among individuals of greater weight. The ability to examine the role of a very extensive list of comorbidities is a major strength of this analysis.
Many of the identified risk factors, while complex, multifactorial, and often not necessarily modifiable, provide an impetus to follow patients at higher risk for readmission more aggressively following discharge. Prolonged length of stay, for example, was identified as an important risk factor yet the reasons for the longer stay varied widely in BOLD; both preoperative and perioperative factors can interact to influence the duration of a patient's stay. Despite this heterogeneity, prolonged length of stay could be utilized as a prompt for enhanced post-discharge monitoring in patients at higher risk for readmission. Intervention studies are needed to determine if and how enhanced monitoring, adjunctive treatments, or additional education might reduce readmission rates for high-risk patients. Certainly, enhanced monitoring is unlikely to prevent more serious readmissions such as those in patients who develop gastrointestinal leaks or obstructions. Further, it is unknown at this time how much effort would need to be applied to significantly lower the current readmission rates that are already acceptably low. However, it may be possible to impact the most prevalent reasons for readmission, nausea and vomiting, by establishing infusion centers for patients suffering from a slow return of bowel function and dehydration.
It is important to recognize the magnitude of relative risk differences associated with the predictors of readmission in this analysis; primarily, comorbid conditions must be weighed with the absolute risk for each procedure. For example, a relative risk of 1.5 for a high-risk patient group compared to a group of typical patients would equate to an increase of the readmission rate from 5.8% to 8.7% for RYGB and from 1.2% to 1.8% for AGB, or absolute risk differences of 2.9% and 0.6%, respectively. Patients, payers, and practitioners alike may find these higher risks acceptable if outweighed by the benefits of surgery, which are often greatest among patients with more severe comorbidity profiles. For these reasons, we deem that the current results do not support patient selection, but rather highlight patient groups that could benefit from appropriate preventative or educational efforts, and possibly, closer post-discharge follow-up.
This study has several limitations. The exclusion of centers with low follow-up rates is the most important limitation, since it is possible that BSCOE excluded for low follow-up rates may have been the hospitals with the highest readmission rates, as well. Our sensitivity analyses comparing included and excluded centers did not indicate that there were substantive clinical differences between the patient populations ( and ), and if such a bias were present, it is highly unlikely that an underestimation of readmission rates would have a considerable impact on the strength or direction of the risk factors themselves. Long-term follow-up in BOLD was limited and precluded the examination of readmissions occurring beyond 30 days. RYGB patients continue to require readmission up to and beyond one year, and the need for band revisions generally do not occur within the first 30 days of surgery; however, the greatest proportion of readmissions occur within 30 days 
, so the current study likely captures the most important risk factors for readmission.
Comparison of RYGB Patients from Included and Excluded BSCOE.
Comparison of AGB Patients from Included and Excluded BSCOE.
The present analysis was also unable to control for surgeon volume, an important factor in readmission 
. however surgeons reporting to BOLD must log over 50 cases annually in order to maintain BSCOE certification. Also, the distribution of readmission rates was not consistent with a uniform rate across centers ( and ). An appreciable number of BSCOE, for AGB in particular, reported readmission rates considerably lower than would be expected under a constant rate across centers of varying surgical volume. We suspect that unusually low BSCOE readmission rates reflect unmeasured variables such as surgical experience, and those with high rates of readmission could be indicative of either surgeon inexperience, case difficulty, or surgeon preference for open procedures. Data entered into BOLD is self-reported by BSCOE, so post-discharge events are potentially underreported in the database, though our selection of centers with high follow-up rates was undertaken to offset potential underreporting in the larger database.
Loess plot of RYGB readmission rates on follow-up rates.
Loess plot of AGB readmission rates on follow-up rates.
Finally, the observational nature of the study precludes causal inference about risk factors. Given that examination of factors influencing relatively rare events like readmissions requires thousands of patients to be adequately powered to assess differences in risk, it is unlikely that randomized studies of these factors will ever be performed. Therefore, decisions on patient selection and risk calculations will inevitably be based on large prospective observational databases like BOLD. Nested case-control studies, in which more extensive collection of possible explanatory variables is performed, may shed light on the problem of unmeasured confounders in the BOLD dataset.
In conclusion, we have characterized patterns of risk for readmission associated with patient and intraoperative factors for the two most common bariatric procedures in the largest prospective cohort of bariatric surgery patients to date. While the overall readmission rates for both procedures are low, the present results may prove to be an important clinical tool in the development of patient education programs, algorithms for procedure selection, and follow-up plans. In an effort to maximize patient benefit and cost-effectiveness of bariatric surgery and to reduce penalties from payers, primary care providers and surgeons should understand patient-specific risks to optimize clinical care for patients when both selecting for and immediately following their bariatric operation.