Comparative Effectiveness of Laparoscopic Adjustable Gastric Banding vs Laparoscopic Gastric Bypass

Importance  Laparoscopic Roux-en-Y gastric bypass (RYGB) and laparoscopic adjustable gastric banding (AGB) are 2 of the most commonly performed bariatric procedures worldwide. However, few large, multisite studies have directly compared the benefits and harms of these procedures.

Objective  To compare the effect of laparoscopic RYGB vs AGB on short- and long-term health outcomes.

Design, Setting, and Participants  A retrospective cohort study of 7457 individuals 21 years or older who underwent laparoscopic bariatric surgery from January 1, 2005, through December 31, 2009, with follow-up through December 31, 2010. All individuals were participants in the Scalable Partnering Network, a network of 10 demographically and geographically distributed health care systems in the United States.

Main Outcomes and Measures  The primary outcomes were (1) change in body mass index (BMI), (2) a composite end point of 30-day rate of major adverse outcomes (death, venous thromboembolism, subsequent intervention, and failure to discharge from the hospital), (3) subsequent hospitalization, and (4) subsequent intervention.

Results  We identified 7457 patients who underwent laparoscopic AGB or RYGB procedures with a median follow-up time of 2.3 years (maximum, 6 years). The mean maximum BMI (calculated as weight in kilograms divided by height in meters squared) loss was 8.0 (95% CI, 7.8-8.3) for AGB patients and 14.8 (95% CI, 14.6-14.9) for RYGB patients (P < .001). In propensity score–adjusted models, the hazard ratio for AGB vs RYGB patients experiencing any 30-day major adverse event was 0.46 (95% CI, 0.27-0.80; P = .006). The hazard ratios comparing AGB vs RYGB patients experiencing subsequent intervention and hospitalization were 3.31 (95% CI, 2.65-4.14; P < .001) and 0.73 (95% CI, 0.61-0.88; P < .001), respectively.

Conclusions and Relevance  In this large bariatric cohort from 10 health care systems, we found that RYGB resulted in much greater weight loss than AGB but had a higher risk of short-term complications and long-term subsequent hospitalizations. On the other hand, RYGB patients had a lower risk of long-term subsequent intervention procedures than AGB patients. Bariatric surgery candidates should be well informed of these benefits and risks when they make their decisions about treatment.

The prevalence of severe obesity, defined as a body mass index (BMI; calculated as weight in kilograms divided by height in meters squared) of 40 or greater, has been increasing more rapidly than mild or moderate obesity in the United States,^1,2 and it is anticipated that 9% of adults will be severely obese by 2030.³ In the last several decades, bariatric surgical procedures have emerged as highly effective treatment options for inducing weight loss and improving comorbidities among adults with severe obesity.⁴ According to the National Institutes of Health, bariatric procedures should be considered for people with a BMI of 40 or greater or those with a BMI of 35 or greater who also have obesity-related comorbidities and have failed to achieve significant weight loss through nonsurgical approaches.⁵ Although many types of bariatric procedures currently exist, the existing evidence base does not clearly support use of one procedure over another.

First introduced in 1994, laparoscopic Roux-en-Y gastric bypass (RYGB) and laparoscopic adjustable gastric banding (AGB) are now 2 of the most commonly performed bariatric procedures worldwide.⁶ However, there are important tradeoffs between the potential risks and benefits of the 2 approaches. There is ongoing debate about whether the AGB and RYGB procedures can achieve comparable weight loss, with systematic reviews^7-10 yielding conflicting results. The RYGB has traditionally been associated with more frequent and serious perioperative complications than the AGB,⁷ although the rates of complications vary widely in the literature. However, the RYGB procedure is also associated with greater remission of comorbid health conditions,^7,9 including type 2 diabetes mellitus, which is believed to be at least partially mediated through mechanisms that are independent of changes in weight.¹¹ Most of the data on the outcomes of the RYGB and AGB procedures come from surgical case series and single-procedure observational studies,^7-10 and there are few comparative studies. A systematic review¹² identified only 2 randomized trials that compared the RYGB with the AGB.

Given the relative paucity of research in this area, the goal of our current study was to better understand the comparative effectiveness of the RYGB vs the AGB on long-term weight loss, as well as short- and long-term complications of each procedure. We hypothesized that the RYGB would have superior weight loss but greater short- and long-term complication rates than the AGB.

This study was approved by the Kaiser Permanente Colorado Institutional Review Board, and the requirement for informed consent was waived. We conducted this retrospective cohort study in 10 health care systems participating in the Scalable Partnering Network (SPAN) for comparative effectiveness research (http://www.span-network.org/),¹³ including Group Health (Washington), Geisinger Health System (Pennsylvania), HealthPartners (Minnesota), Harvard Pilgrim (Massachusetts), Essentia Institute of Rural Health (Minnesota), and Kaiser Permanente regions in Colorado, Georgia, Hawaii, Northern California, and the Northwest (Oregon and Washington). The 10-site SPAN bariatric database was constructed using individual-level patient data extracted via common programming language from electronic medical records, pharmacy databases, and insurance claims. Individuals were included in this study if they underwent primary laparoscopic RYGB or AGB surgery from January 1, 2005, through December 31, 2009, were 21 years or older at the time of initial bariatric surgery, and were enrolled in the health care system the year before surgery. Individuals were excluded if all recorded BMIs were less than 35 during the presurgery year. All (except 2 of the 4 surgical centers associated with Kaiser Permanente Northern California) were certified as Centers of Excellence or Accredited Bariatric Surgery Centers with the American College of Surgeons or the American Society of Metabolic and Bariatric Surgery/Surgical Review Corporation.

The exposure was a first laparoscopic RYGB or laparoscopic AGB procedure that was identified via Healthcare Common Procedure Coding System codes, International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM), and Current Procedure Terminology codes (the full list of codes is in eTable 1 in the Supplement).

The primary outcomes were (1) a composite end point of 30-day rate of major adverse outcomes, (2) the rate of subsequent hospitalization, (3) the rate of subsequent intervention, and (4) the mean maximum change in BMI from baseline. The composite end point of 30-day major adverse outcomes was based on the definition used in the Longitudinal Assessment of Bariatric Surgery (LABS) study and included death; venous thromboembolism; percutaneous, endoscopic, or operative subsequent intervention; and failure to be discharged from the hospital.¹⁴ Deep vein thrombosis and pulmonary embolism were identified using ICD-9-CM diagnosis codes (eTable 1 in the Supplement).

Consistent with LABS study definitions, subsequent interventions in our composite 30-day outcome and long-term outcomes were defined as any additional bariatric procedure and other procedures related to device removals, gastric revisions, abdominal or incisional hernia repair, laparoscopy or laparotomy, and percutaneous endoscopic gastrostomy tube placements.¹⁴ We excluded gastric endoscopic procedures from our primary subsequent intervention definition; however, sensitivity analyses that included the endoscopic procedures are presented in eFigure 1 and eTable 2 in the Supplement. Long-term subsequent interventions were considered from the bariatric procedure surgery date until the censoring date, which was defined as the earliest date of one of the following: death, end of continuous membership in the health care system, or end of the study period (December 31, 2009, which was the last day reliable mortality data were available).

All incident hospitalizations from the bariatric procedure hospital discharge date until the censoring date, as defined in the previous paragraph, were eligible for our analyses of subsequent hospitalization. Although hospitalizations in the first year after bariatric surgery are more likely to be related to the initial procedure than hospitalizations that occurred during or after the second postoperative year, we sought to examine the long-term association between the RYGB vs AGB procedures and the need for any inpatient care. We also did not adjudicate the cause of hospitalization through medical record reviews.

Finally, we examined change in body weight among the subpopulation of bariatric patients who had at least 2 valid BMI measurements within 2 years after the index surgery. Follow-up for BMI measurements commenced at the bariatric surgery procedure date and ended at the earliest date of death, end of continuous membership, the first evidence of pregnancy (determined by diagnosis and procedure codes), or the end of the BMI outcome study period on December 31, 2010.

To handle potential confounding between the effects of the RYGB vs AGB procedures on our outcomes, survival models were adjusted using strata defined by propensity score quintiles.¹⁵ To calculate propensity scores, we constructed a logistic regression model of index surgery that included the following covariates for all models: health care plan site, year of index procedure, race/ethnicity, smoking status, sex, Charlson Comorbidity Index score, 11 individual comorbidities (based on ICD-9-CM diagnosis codes for atrial fibrillation, diabetes mellitus, gastroesophageal reflux, hypertension, sleep apnea, asthma, deep vein thromboembolism, pulmonary embolism, congestive heart failure, dyslipidemia, and coronary artery disease, all within the year before the index surgery), supplemental oxygen use, assistive walking device use, insurance type (Medicaid, Medicare, commercial, or other), drug coverage before index surgery, anticoagulant dispensing before surgery, number of days between baseline BMI and surgery, most recent BMI before surgery, length of presurgical plan enrollment, age, most recent systolic and diastolic blood pressure (BP), number of days before surgery for BP measurement, missing BP value, total length of stay for all inpatient hospitalizations within 1 year before index surgery, and demographic variables estimated from US Census information for each patient’s zip code (median family income, educational level less than high school, and missing US Census data). We used stepwise variable selection with a significance to enter and stay set at 10% to determine which covariates to include. Balance was checked by comparing surgery proportions by levels of categorical covariates within propensity score strata and by comparing means of continuous covariates by surgery within strata.

Missing values were present in the median family income, educational level, timing for presurgical BP, diastolic BP, and systolic BP continuous covariates. Because the rate of missing value records was low (≤2%) for these covariates, missing values were singly imputed using the mean of the nonmissing values, and a comparison with complete-case analysis was conducted to determine whether the results changed. Missing values were also present in categorical variables for race/ethnicity and smoking status, for which an additional category was created.

Cox proportional hazards survival modeling was used to estimate the hazard ratios (HRs) comparing AGB and RYGB operations on the 30-day composite and long-term adverse event outcomes (subsequent hospitalization and intervention). The main models included surgery type as the independent variable and were stratified on propensity score quintiles. To ensure adequate adjustment for covariates, additional models were estimated, including site indicators, patient characteristics, interactions between index surgery type and site, propensity score quintiles, and combinations of these effects as independent variables. Nonstratified and site-stratified models were also estimated. The surgery HRs and SEs were then compared among all models for each outcome. To check the sensitivity of the models to the method of handling missing values, the main models were also estimated after excluding patients with missing US Census and BP data. The proportional hazards assumption was also assessed for the main models using the Kolmogorov-Smirnov supremum test and visual inspection of cumulative sums of Martingale residuals.¹⁶ We conducted 2 sensitivity analyses for all survival models in 2 extreme scenarios: (1) assuming those who disenrolled immediately had an event and (2) assuming those who disenrolled would never have experienced the event through the maximum follow-up time. Our study results were qualitatively similar with these alternate specifications.

To examine changes in BMI over time, we used an inverse intensity-of-visit process–weighted generalized estimating equation model for normally distributed data to mitigate bias due to potential correlation among the number, timing, and values of the weight measurements.¹⁷ Separate B-spline basis functions were created for each surgery and included in the model to capture the nonlinearity of BMI change trends over time. All covariates previously listed for the propensity model were included here, and missing values for continuous variables were imputed similarly. A plot of the modeled BMI trend over time and summary measures that compared weight change between and within the AGB and RYGB procedures were created from the summarization of estimates from 200 bootstrap replicates of this model. For the modeled trend and estimates of weight change by surgery, model covariates were set to the median values for continuous variables and the mode for categorical variables. All statistical analyses were conducted using SAS statistical software, version 9.3 (SAS Institute Inc).

We initially identified 12 111 patients 21 years or older who had undergone a laparoscopic AGB or RYGB procedure from January 1, 2005, through December 31, 2009, and had been enrolled in the health care system for 1 year before the procedure (eFigure 2 in the Supplement). Among these, we had to exclude 4377 patients because they did not have a preoperative BMI value in the electronic medical databases because they were cared for in nonintegrated settings. An additional 277 patients were excluded because they lacked 2 or more BMI measures in the first 2 years after their procedure for estimating changes in BMI over time. Therefore, the final study population for the BMI outcome consisted of 7457 individuals across 10 sites who met inclusion and exclusion criteria and for whom available data were sufficient (eFigure 2 in the Supplement). Patients with missing BMI data were similar to those with available BMI data with respect to sex, surgery type, and number of comorbidities; however, patients with missing BMI data were slightly younger (43.7 vs 46.0 years). The study population for the 30-day composite adverse outcome and long-term subsequent hospitalization and intervention outcomes consisted of 6992 individuals. This smaller cohort size for the adverse event outcomes was a result of the need to exclude patients from one health care system that was unable to report mortality data.

The Table provides descriptive information on the main study population. Overall, the RYGB procedures were more common than the AGB procedures (5950 RYGB procedures vs 1507 AGB procedures). There were many significant demographic and clinical differences between patients who underwent the RYGB and AGB procedures (Table).

Among the 6992 patients who were eligible for our analyses of short-term composite adverse events, 6643 (95.0%) had event-free follow-up through 30 days after their initial bariatric procedure (5497 RYGB patients [94.8%] vs 1146 AGB patients [96.1%]), 189 (2.7%) experienced one of the composite adverse events, 41 (0.6%) disenrolled from their health care plan, and 119 (1.7%) were censored because the study period ended. A significantly greater proportion of the RYGB patients experienced 1 or more major adverse events by 30 days than the AGB patients (eTable 3 in the Supplement; 174 RYGB patients [3.0%] vs 15 AGB patients [1.3%]; P < .001). Of the 174 RYGB patients who experienced a major adverse event, 1 died (0.6%) and 8 were still hospitalized at 30 days (4.6%); 112 patients had a subsequent intervention (64.4%) and 62 patients had a deep vein thrombosis or pulmonary embolism (35.6%). Among the 15 AGB patients who experienced a major adverse event, none died or were still hospitalized at 30 days; however, 9 (60.0%) had a subsequent intervention and 6 (40.0%) had a deep vein thrombosis or pulmonary embolism. In the unadjusted model, the HR for the AGB vs RYGB patients experiencing any 30-day major adverse outcome was 0.42 (95% CI, 0.25-0.71; P = .001). In the propensity score–adjusted model, the HR was 0.46 (95% CI, 0.27-0.80; P = .006). Figure 1 shows the unadjusted cumulative incidence of major adverse events within 30 days by procedure type.

The median follow-up time was 1.5 years (1.56 years for the RYGB patients vs 1.29 years for the AGB patients; P < .001) for the 6992 patients who were eligible for our analyses of long-term subsequent intervention and hospitalization outcomes. The maximum follow-up time was 5.0 years, during which 1064 patients (15.2%) were lost to follow-up because of disenrollment from their health plan. Only 525 patients (7.5%) were lost to follow-up because of disenrollment at the end of year 1. During the entire follow-up period of 1192 AGB patients, 2 patients (0.2%) died, 148 patients (12.4%) were hospitalized again, and 163 patients (13.7%) had 1 or more subsequent interventions in the AGB cohort (eTable 2 in the Supplement). In the RYGB cohort of 5800 patients, 17 patients (0.3%) died, 1155 patients (19.9%) were hospitalized again, and 318 patients (5.5%) had 1 or more subsequent interventions. The most common subsequent interventions for the AGB patients were additional abdominal surgical procedures, which occurred in 107 (9.0%) of the 1192 AGB patients (eTable 2 in the Supplement). Removal of the band occurred in 26 (2.2%) of 1192 AGB patients. The most common subsequent interventions for RYGB patients were additional abdominal surgical procedures, occurring in 157 patients (2.7%) (eTable 2 in the Supplement).

In the unadjusted model, the HR comparing the AGB and RYGB patients experiencing a long-term subsequent intervention was 3.15 (95% CI, 2.60-3.81; P < .001), favoring the RYGB procedure. In the propensity score–adjusted model, the HR was 3.31 (95% CI, 2.65-4.14; P < .001). The unadjusted and propensity score–adjusted HRs comparing the AGB to RYGB patients for experiencing a subsequent hospitalization were 0.70 (95% CI, 0.59-0.84; P < .001) and 0.73 (95% CI, 0.61-0.88; P < .001), respectively, favoring the AGB procedure. Figure 2 shows the long-term, unadjusted cumulative incidence of subsequent intervention comparing the AGB and RYGB procedures, and Figure 3 shows the long-term, unadjusted cumulative incidence of subsequent hospitalization comparing the AGB and RYGB procedures.

Among the 7457 patients who were eligible for our BMI change analyses, the median follow-up time was 2.3 years, and the maximum follow-up time was 6.0 years, with 91.5%, 59.8%, and 33.7% having 1, 2, and 3 years of follow-up for analysis. During the entire study period, 1456 (19.5%) disenrolled (were lost to follow-up), but only 533 (7.1%) were lost to follow-up at the end of year 1. During this time, the mean maximum BMI loss was 8.0 (95% CI, 7.8-8.3) for the AGB patients and 14.8 (95% CI, 14.6-14.9) for the RYGB patients (P < .001 for the RYGB vs AGB comparison). The estimated BMI loss at 1, 2, and 3 years was 6.4 (95% CI, 6.0-6.7), 6.2 (95% CI, 5.8-6.7), and 5.3 (95% CI, 4.7-5.9) for the AGB patients and 13.9 (95% CI, 13.7-14.2), 13.5 (95% CI, 13.2-13.8), and 12.1 (95% CI, 11.8-12.4) for the RYGB patients. The RYGB patients had 7.6 (95% CI, 7.3-7.9), 7.3 (95% CI, 6.8-7.7), and 6.8 (95% CI, 6.2-7.4) greater decreases in BMI than the AGB patients at 1, 2, and 3 years, respectively. Comparisons of BMI loss between operations were statistically significant at all time points (P < .001). Figure 4 illustrates the multivariable adjusted, inverse intensity–weighted trajectories of change in BMI for cohorts undergoing each procedure.

In this large retrospective cohort study of 10 geographically diverse health care systems in the United States, we found compelling evidence that the AGB and RYGB procedures have different short- and long-term effects on health outcomes. The AGB procedure had a significantly lower 30-day risk of major adverse events than the RYGB procedure (1.3% vs 3.0%; HR, 0.46). This translates into a number needed to treat of 58 patients undergoing AGB instead of RYGB to prevent one major adverse event. Despite the lower 30-day risk, we found that the AGB patients had a significantly greater risk of subsequent intervention than the RYGB patients during a median of 1.5 years (HR, 3.31). Interestingly, the AGB cohort had a 27% lower risk of subsequent hospitalization during the same period (HR, 0.73), owing to a relatively high (19.9%) rate of subsequent hospitalization among the RYGB patients. It is also possible that the subsequent hospitalization rate is relatively high in the RYGB cohort because subsequent interventions in the AGB cohort were more likely to occur as outpatient procedures. Finally, although we observed major reductions in the BMI within the first year for both cohorts, we found that the RYGB patients experience nearly double the weight loss of the AGB patients at 1 to 3 years after surgery. Beyond 2 years, weight regain was observed in both groups.

The 30-day composite adverse event rates (death, venous thromboembolism, subsequent intervention, or failure to discharge from the hospital) that we report are similar to the LABS study for the AGB procedure but slightly lower than what has been previously reported for the RYGB procedure. The LABS study observed a 30-day composite event rate of 1.0% among 1198 laparoscopic AGB patients and a 4.8% rate among 2975 laparoscopic RYGB patients.¹⁴ It appears that this difference is attributable to the fact that the RYGB cohort in the LABS study had a higher rate of operative subsequent intervention (3.2% vs 1.9%) than our population.

The findings for weight loss reported are consistent with much of the prior literature.^{4,7,9,12,18,19} In a meta-analysis of 1959 AGB patients and 2705 RYGB patients, Buchwald and colleagues⁹ found a mean BMI decrease of 10.8 for the AGB patients and 17.1 for the RYGB patients. Our estimates for maximum BMI loss after surgery were similar, albeit slightly lower, at 8.0 for the AGB patients and 14.8 for the RYGB patients. Another systematic review⁸ of 19 long-term studies (≥10 years’ duration; no randomized trials) is at odds with our findings and suggests that long-term weight losses are similar between the AGB and RYGB patients (mean estimated weight loss, 54.2% for the AGB patients vs 54.0% for the RYGB patients). These discrepant data suggest that some experienced high-volume surgical centers with rigorous programs for long-term postsurgical care and follow-up may achieve weight loss results with the AGB patients that approximate those achieved with the RYGB patients. Our results may better reflect expected findings in a broad community-based population. Further research in all settings is needed to examine the durability of weight loss for both procedures, as well as predictors of long-term weight regain.

Although the AGB procedure appears to have a more favorable short-term risk profile, other long-term studies^18,19 have suggested that rates of band removal in the AGB patients are likely to be much higher than the 2% reported here and may be as high as 50%. Removal of bands in the AGB patients is typically because of failure to achieve or maintain clinically significant weight loss; band malfunction, slippage, or erosion; or patient intolerance of the gastric restriction.

Strengths of our study include our cohort of more than 7000 geographically and demographically diverse individuals with well-defined presurgical and postsurgical clinical data that characterize a spectrum of beneficial and adverse outcomes. Other strengths include applying criteria for categorizing major adverse events that are consistent with the LABS study.¹⁴ Our study also has some important limitations. The long-term outcome of subsequent interventions was a broad category, including procedures ranging from incisional hernia repair to therapeutic laparotomy (eTable 2 and eFigure 1 in the Supplement), and the indications for these procedures were not adjudicated by formal medical record review. As such, this outcome may overestimate the number of subsequent interventions that are directly attributable to the AGB and RYGB procedures. We also did not have the resources to adjudicate the reasons for each subsequent hospitalization. Despite these limitations in the causal link, our analyses accurately reflect true differences in health care resource use across these 2 groups of patients, and the reported differences in rates of procedures and hospitalizations are likely to be informative to patients, physicians, and policymakers. Another limitation was the lack of preoperative BMI data on 4377 patients who had a bariatric procedure performed in clinical locations that lacked electronic medical records to capture those data; however, we do not believe this mechanism for missing data is likely to introduce any bias into our study because we found no meaningful differences in the characteristics of patients with and without BMI data.

We found important differences in short- and long-term health outcomes for the AGB and RYGB procedures across 10 health care systems in the United States. Severely obese patients should be well informed of these differences when they make their decisions about treatment.

Accepted for Publication: April 4, 2014.

Corresponding Author: David Arterburn, MD, MPH, Group Health Research Institute, 1730 Minor Ave, Ste 1600, Seattle, WA 98108 (arterburn.d@ghc.org).

Published Online: October 29, 2014. doi:10.1001/jamasurg.2014.1674.

Author Contributions: Dr Bayliss had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Arterburn, Polsky, Butler, Donahoo, Bayliss.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Arterburn, Portz, Bayliss.

Critical revision of the manuscript for important intellectual content: Arterburn, Powers, Toh, Polsky, Butler, Donahoo, Herrinton, Williams, Vijayadeva, Fisher, Bayliss.

Statistical analysis: Powers, Toh, Butler.

Obtained funding: Bayliss.

Administrative, technical, or material support: Arterburn, Polsky, Butler, Portz, Donahoo, Herrinton, Williams.

Study supervision: Arterburn, Herrinton, Vijayadeva, Bayliss.

Conflict of Interest Disclosures: Dr Herrinton reported receiving funding from Centocor, Procter & Gamble, Genentech, and Medimmune. No other disclosures were reported.

Funding/Support: This project was supported by grant 1R01HS019912 from the Agency for Healthcare Research and Quality.

Role of the Funder/Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and the decision to submit the manuscript for publication.

Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the Agency for Healthcare Research and Quality.

Additional Contributions: Programmers/analysts from the following SPAN sites provided data for this project: Denver Health and Hospital Authority (DHHA), Essentia Institute of Rural Health (EIRH), Geisinger Health System (GHS), Group Health Research Institute (GHRI), Harvard Pilgrim Healthcare (HPHC), HealthPartners Institute for Education and Research (HPIER), Kaiser Permanente Hawaii (KPHI), Kaiser Permanente Northern California (KPNC), Kaiser Permanente Northwest (KPNW), Kaiser Permanente Colorado (KPCO), and Kaiser Permanente Georgia (KPGA). We are indebted to the following SPAN investigators for the collaborations that made this study possible: KPCO (Matthew F. Daley, MD; Ella Lyons, MS; Marsha A. Raebel, PharmD; Lisa Pieper, MSHA, MBA), KPNC (Ameena T. Ahmed, MD, MPH), EIRH (Thomas E. Elliott, MD); GHS (Jove Graham, PhD; Christopher Still, DO; Craig Wood, MS), GHRI (Denise Boudreau, PhD, RPh); HPIER (Pamela Pawloski, PharmD); KPHI (Cynthia Nakasato, MD), DHHA (Art Davidson, MD); and KPNW (Lynn DeBar, PhD, MPH; Kristine Funk, MS, RD). All individuals listed received support from grant 1R01HS019912 from the Agency for Healthcare Research and Quality for this project. Michael Shainline, MS, provided technical and administrative assistance.