An official website of the United States government
NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.
Amir A. Ghaferi, MD, MS, Sarah T. Hawley, PhD, MPH, Angela Fagerlin, PhD, Mousumi Banerjee, PhD, Lawrence An, MD, Lisa A. Prosser, PhD, and Nancy Birkmeyer, PhD.
Author Information and AffiliationsTreatment options for morbid obesity include medical management (diet, exercise, behavior modification, weight loss medications, etc) and 1 of 2 main types of bariatric surgery: Roux-en-Y gastric bypass or sleeve gastrectomy. The risks and benefits of these options vary widely and are affected by patient and clinical characteristics. Patients may struggle with deciding which intervention is best for their individual circumstance. Shared decision-making should reflect informed patient values and preferences regarding these trade-offs.
The goals of this research were to develop, implement, and evaluate an informed decision support tool for the treatment of morbid obesity.
We developed a web-based interactive decision support tool that includes (1) a discrete-choice experiment (conjoint analysis, a survey-based approach for measuring preferences2) to identify patient preferences for risks, benefits, and other attributes of the treatment options; (2) tailored information regarding the risks and benefits of the treatment options; and (3) information regarding other salient attributes of the treatment options. We then conducted a randomized controlled trial (RCT) comparing the decision support tool with usual care in patients having surgery at 38 hospitals in Michigan. Primary outcomes included decisional measures (treatment choice, preference concordance, and decisional conflict) and clinical patient outcomes (weight loss, patient satisfaction, and quality of life). Secondary outcomes included satisfaction with and usage data from the decision tool and the patients who used it. We used these data to create and test a revised tool and using a propensity-matched analysis.
After creation of the phase 1 decision tool, we randomly assigned 878 patients into the study protocol (control, n = 440; intervention, n = 438). Of those who were randomly assigned, 815 patients (92.8%) completed the conjoint analysis. In these analyses, participants choose between hypothetical profiles composed of attributes of bariatric surgery procedures. Preferences for attributes are expressed as coefficient values. A value >0 indicates that including that attribute level in a surgery profile makes it more likely that the participant will choose that profile. Patients were most likely to select profiles that included resolution of medical conditions (coefficient for full resolution = 0.229 [95% CI, 0.177-0.280], P < .001; coefficient for no resolution = −0.207 [95% CI, −0.284 to −0.189], P < .001), higher total weight loss (coefficient for each additional 20% loss = 0.185 [95% CI, 0.166-0.205], P < .001), and lower out-of-pocket costs (coefficient for each additional $1000 = −0.034 [95% CI, −0.042 to −0.025], P < .001).
After randomization, similar proportions underwent bariatric surgery, with 35% in the phase 1 decision tool study group and 38% in the control group. The 2 groups had similar proportions of the bariatric surgery procedures. The decision tool did not affect patient decisional outcomes (ie, overall decisional regret, 80.9 ± 21.5 in the control group vs 83.8 ± 17.5 in the intervention group, P = .20; the scale was 0-100, with lower scores indicating greater regret), satisfaction with surgery at 1 year (79.6% of control patients were very satisfied, and 85.1% of intervention patients were very satisfied), or clinical outcomes (ie, percentage of excess weight lost at 1 year, 61.0% ± 19.8% in the control group vs 61.1% ± 18.3% in the intervention group).
We then conducted a propensity-matched comparison of the 154 patients who completed the revised tool and had surgery with 159 RCT control group patients who had surgery. Patients who received the revised tool had less decisional regret than did propensity-matched control patients, with a mean difference of 7.1 points (95% CI, 1.6-12.7 points; P = .0118).
The decision to choose one weight loss method over another is complex. This study provides information on which characteristics are most important from a patient perspective and shows that preferences are heterogeneous. Further, presenting patients with personalized outcomes and preference data does not necessarily reduce decisional regret or affect clinical outcomes.
This study had 4 main limitations. First, the main analysis of the RCT was a per-protocol analysis in the subgroup of patients who underwent bariatric surgery, which could introduce selection bias. Second, the tool was developed and discussed at the statewide collaborative's triannual meetings, where providers may have gained knowledge that would influence their discussions with patients and potentially affect the outcomes and satisfaction of our control patients. Third, a post hoc study of the revised decision tool uses historical controls; the results support conducting a future RCT to confirm the findings. Fourth, the timing of recruitment led to a lower-than-anticipated rate of patients following through with bariatric surgery.
Obesity is increasingly considered among the most important public health problems of our times. In the United States, two-thirds of adults are now overweight or obese overall. During the last 2 decades, the most rapid increases have been in the prevalence of morbid obesity. At least 15 million American adults are morbidly obese (body mass index [BMI] >40, or >35 with obesity-related comorbidity), a number that has more than quadrupled in the last 20 years.3
Bariatric surgery is the only treatment that has proven effective in producing long-term weight loss for patients with morbid obesity.4,5 Bariatric surgery also results in the resolution of obesity-related comorbid conditions, improvements in quality of life (QOL), and increased survival. There are currently 4 different bariatric surgical procedures in use: (1) adjustable gastric banding and (2) duodenal switch, and the more prominent (3) Roux-en-Y gastric bypass (RYGB) and (4) sleeve gastrectomy (Table 1).
Explanation of Bariatric Surgical Options.
The risks and benefits of bariatric procedures are influenced by independent risk factors for serious complications, such as baseline BMI, age, male sex, history of smoking, and mobility limitations.1 Herein, we present data from the Michigan Bariatric Surgery Collaborative (MBSC), a consortium of approximately 40 hospitals and 80 surgeons focused on quality measurement and improvement. The patient and procedure mix in Michigan is very similar to that in the rest of the United States.6 The MBSC provides a unique platform that prospectively collects data and conducts rigorous audits of the entered data. After adjusting for risk and sociodemographic differences, average 30-day rates of complications range from 2.4% for adjustable gastric banding to 15.2% for duodenal switch, and serious complications range from 0.8% for adjustable gastric banding to 6.11% for duodenal switch (Figure 1). On the other hand, average body weight loss 3 years after surgery is just 56 pounds (276 pounds at baseline to 220 pounds at 3 years) for adjustable gastric banding, compared with 139 pounds (316 pounds at baseline to 177 pounds at 3 years) for duodenal switch (Figure 2).
Adjusted 30-Day Rates of Bariatric Surgery Complications (MBSC Data 2007-2017, N = 73 172).
Average Body Weight Over Time by Bariatric Procedure Type (MBSC Data 2007-2017, N = 73 172).
Patients must also weigh other attributes of bariatric procedures before making a final decision. For example, postbariatric diets differ according to the type of procedure. Although postsurgical diets are similar in that the amount of food that a patient can eat at 1 time is restricted, RYGB patients additionally must limit high-sugar and high-fat foods to avoid “dumping syndrome,” which causes nausea, vomiting, dizziness, sweating, and diarrhea. Because they limit food absorption, patients undergoing RYGB and duodenal switch procedures require lifelong vitamin and mineral supplements and blood tests to avoid potentially serious nutritional deficiencies. Patients undergoing sleeve gastrectomy must consume small, frequent meals along with protein supplements to ensure adequate nutrition in the months following surgery. Patients undergoing adjustable gastric band procedures require ongoing doctor visits for adjustments of the band when the rate of weight loss plateaus. Although there is little quantitative information on the extent to which patient utility for these aspects of bariatric surgical procedures varies, attendance at any bariatric support group or chat room would indicate that they significantly impact decision-making and QOL.7
Bariatric surgeons also have preferences about the different bariatric procedures (Figure 3). While some bariatric surgeons perform only 1 type of procedure (either RYGB or adjustable gastric band), most surgeons perform 2 procedures: adjustable sleeve gastrectomy and RYGB. Only a small number of surgeons perform adjustable gastric band or duodenal switch procedures. Even among the majority of surgeons who perform multiple different types of procedures, the distribution of the procedure mix varies, especially between the RYGB and sleeve gastrectomy procedures.
Distribution of Bariatric Procedure Mix by Surgeon in Michigan in 2017.
Previously developed decision aids for bariatric surgery have 2 significant limitations. First, they are often out of date and do not contain information on the most prevalent types of surgery. For example, the distribution of procedures performed across Michigan and the United States has shifted predominantly to sleeve gastrectomy and RYGB. A popular tool from the Foundation for Medical Decision Making did not include any information on sleeve gastrectomy as a standalone procedure.8 Second, existing decision aids often provide average risks of procedures for patients to evaluate. They are not customized to the individual patient's age, sex, race, BMI, and comorbidities. Providing a customized, up-to-date tool that can be updated with contemporary procedures is a significant gap in this important area of patient decision-making.
The goals of this research proposal were to develop, implement, and evaluate an informed decision support tool for the treatment of morbid obesity. This decision support tool provides patients with customized estimates of the risks and benefits of different treatment options and also information about other attributes of the options that affect decision-making.
This research promotes shared medical decision-making for patients who are considering bariatric surgery for the treatment of morbid obesity. It helps improve decision quality and outcomes of care for patients. It may also result in improved efficiency of care to the extent that it serves to augment or guide communication between the patient and physician to promote shared medical decision-making.
Our specific aims were as follows:
The stakeholders who play a critical role in providing care, advocacy, and education for those with morbid obesity include providers (physicians, nurses, dieticians, etc), payers, postbariatric-patient caregivers, and professional organizations. To obtain a broad yet appropriately sized sample of this community, we chose 5 key stakeholder groups (see Table 2). These include providers from a state-based bariatric surgery clinical registry, prior bariatric surgery patients, bariatric surgery patient caregivers, representatives from a large insurance company, and “patient experts” (previous bariatric surgery patients who are also professionally affiliated with various bariatric surgery programs across Michigan).
Stakeholder Groups and Perspectives.
We used 3 primary methods for maintaining collaboration with these stakeholder groups during this study. First, the principal investigator, study team members, and key stakeholder representatives met twice during 2014 for our physician/payer and expert stakeholder panel/patient/caregiver kickoff meetings. Second, the study team and various stakeholder representatives met regularly (monthly) both in person and via teleconference during phases 1 and 2 (qualitative work and decision tool development) for the first iteration of the tool and during phase 4 (decision tool usability assessment) for the revised version of the tool. Finally, we provided triannual tool demonstrations and updates to representatives from all stakeholder groups during the MBSC meetings, which took place every February, June, and October from January 1, 2014, to June 30, 2018. In the following, we describe our contacts with each of our stakeholder groups, followed by a table outlining the ways in which our collaborations have improved our decision tool.
On February 7, 2014, we convened our kick-off expert stakeholder meeting, which included the provider and payer stakeholders described in Table 2. The main goals of this meeting were to (1) draft an outline of attributes for the treatment options to inform a format of guiding questions for subsequent bariatric surgery patient focus groups (described in the Methods section) and (2) solicit input on integrating the tool into different delivery systems and practice settings. We also elicited physician and payer perspectives on topics such as risks and benefits patients should consider when choosing a bariatric procedure; QOL issues affected by bariatric surgery; long-term complications; and how physicians tailor their conversations with potential patients, determine the motivating factors behind a patient's decision to have bariatric surgery, and manage a patient who chooses a bariatric procedure with which they do not agree.
On March 14, 2014, we again engaged our stakeholders to review an initial decision tool outline and conjoint analysis survey. An additional goal of this expert panel was to develop a list of attributes that are important to patients when considering bariatric surgery.
To further refine this list of attributes, we held 12 patient focus groups with additional patients in Detroit, Grand Rapids, and Traverse City, all in Michigan. Focus group participants (n = 93) were recruited from MBSC participating sites and had either participated in a medical weight loss program or had adjustable gastric banding, RYGB, sleeve gastrectomy, or duodenal switch procedures. The focus groups were 1 hour in duration, organized by procedure type, and led by a professional focus group moderator. These group sessions were recorded and transcribed. Focus groups participants were given a $100 incentive. A copy of the discussion guide used to augment these groups can be found in Appendix A.
The study team and various stakeholder representatives met regularly from February to December 2014 (development of first iteration of tool) and from March to October 2017 (development of second iteration of tool). During these times, the information gathered during the aforementioned panels and other patient focus groups (described in the “Methods” section for Aim 1) was used to create our decision tool. The wide variety of expertise of both our study team and stakeholder groups allowed us to create a tool that includes a broad range of perspectives regarding the mix of procedures, practice settings, geography, and patient gender and racial/ethnic diversity.
Progress updates and decision tool demonstrations were provided triannually at 14 MBSC meetings between February 2014 and June 2018. The meeting attendees represented all stakeholder groups. During these meetings, we gathered invaluable feedback and suggestions for tool improvement regarding tool comprehension, usability, and applicability in various clinical settings.
Our stakeholders were instrumental in formulating our research questions, defining essential characteristics and attributes of our decision tool and its measurable outcomes, monitoring study progress, and helping us determine ways to make our tool most effective. Table 3 lists our stakeholder groups, the primary input they provided, and how their input shaped our study design.
Stakeholders and Their Input in the Study Design.
The goals of this intervention were to develop, implement, and evaluate an informed decision support tool for the treatment of morbid obesity per the project phases found in Figure 4. This decision support tool provides patients with customized estimates of their potential risks and benefits of different treatment options as well as information about other attributes of the options that affect decision-making.
Overview of Project Phases.
The conceptual model for this study is informed by the Ottawa Decision Support Framework, which asserts that decision support can improve decision quality by addressing unmet decisional needs.9 Decisional needs include knowledge, expectations, values, support, and resources and may be affected by personal and clinical characteristics. Decision support clarifies decisional needs, provides facts, clarifies values, and provides guidance for deliberation and communication leading to high-quality decisions that are informed and values based.
For our project, decisional needs include information regarding the risks, benefits, and other considerations for each of the treatment options presented in a standardized, understandable way that helps patients clarify their values and preferences. Our decision support tool provides tailored risk/benefit information to patients and their physicians, as well as facts regarding other considerations for each of the treatment options. This information helps enhance and guide patient-physician communication and deliberation to promote shared decision-making and improved patient outcomes.
The first component of our study used a conjoint analysis, a survey-based approach for measuring preferences.2 The premise of conjoint analysis, originally developed for the marketing and transportation sectors, is that any good or service can be described by its characteristics or attributes, and the extent to which an individual values a particular attribute can be quantified and compared with other attributes.10 This is especially useful when evaluating trade-offs between competing options. For example, when designing a motor vehicle, a conjoint analysis can help understand the relative preference for the appearance of a vehicle with the need for increased cargo space. The strength of conjoint analysis is that it offers a quantitative assessment of the attributes of a good or service that are most valued. At the same time, the results are not designed to easily compare existing products as a whole. Therefore, the results of this study allow us to compare the key aspects of 2 surgical procedures but not to compare the procedures themselves. Studies using conjoint analysis are increasingly used in health services research to measure preferences for health interventions and public health programs.11 This method is often used to establish patient, provider, or public priorities for medical care, such as determining which factors parents prioritize when selecting a pediatric medical home.12
We used a discrete-choice experiment, a type of conjoint analysis, which asks respondents to choose between 2 or more alternatives based on their component characteristics.13 Alternatives presented to respondents were hypothetical surgical and medical weight loss profiles, such as greater weight loss but a higher rate of complications. By varying the characteristics of competing profiles, this type of study can be used to determine the characteristics most important to a respondent's choice.
Key research questions for this portion of the study were (1) Which characteristics of bariatric surgical options are most important to potential patients? and (2) What is the relative value of their importance?
Based on the focus groups described in the Patient and Stakeholder Engagement section, 9 attributes of bariatric surgical procedures were included in the final survey (Table 4). Levels for the weight loss attribute were expressed in terms of the percentage of excess weight loss. The respondent-specific excess weight loss was calculated using the respondent's reported BMI and included in the profile descriptions to improve understanding of the weight loss levels. Attributes were defined to reflect a range that would encompass the 4 available bariatric surgical options and medical weight loss, as well as include plausible attributes for procedures that may arise in the future.
Description of Attributes and Levels.
We developed a tailored, interactive decision aid to support patient deliberation and patient-provider communication regarding treatments for morbid obesity based on the conceptual model shown in Figure 5. The development of our decision aid included both the quantitative and qualitative work described below.
Conceptual Model for Improving Patient Decisions about Bariatric Surgery.
In a separate non-PCORI analysis based on the 80 000 bariatric surgery patients already in the MBSC registry, we developed models to predict complications (overall, serious, and fatal); weight loss at 1, 2, and 3 years following surgery; and resolution of various comorbidities, such as diabetes, hypertension, and sleep apnea. Calibration and discrimination of these models are to be evaluated. These models have been used to develop an interactive, risk/benefit calculator app (“MBSC – Weigh the Odds”) (Figure 6). After the input of data on the independent risk factors listed previously, the risk calculator produces a table of the patient's predicted risks and benefits from each type of bariatric procedure. In its current form, the risk/benefit calculator is meant for clinician use rather than patient use. Although they were developed separately, the models from the clinician-facing app were used to inform shared decision-making during the clinical encounters we studied.
Screenshots of Existing “MBSC – Weigh the Odds” Risk/Benefit Calculator App.
To ensure that our decision aid included other information that patients with morbid obesity and their physicians feel is important to decision-making, we leveraged the expertise of our study team members, stakeholders, and prior bariatric surgery patients. In addition to the stakeholder meetings described in the Patient and Stakeholder Engagement section, we also convened 12 patient focus groups held in Detroit, Grand Rapids, and Traverse City, all in Michigan. The focus group participants (N = 93) had either participated in a medical weight loss program or had undergone a bariatric surgical procedure. Separate focus groups were held for each of the treatment options and included patients from a broad range of perspectives regarding geography, gender, and racial/ethnic diversity.
Data from these focus groups were used to develop and refine draft attributes and attribute levels and to rank the importance of the attributes. In addition, these focus groups helped develop the formats used to best communicate the risks and benefits and other attributes of the treatment options.
An experienced moderator conducted each focus group and followed guiding questions developed in the stakeholder panels detailed in the engagement plan. Each group was audio- and video-recorded with participant permission, and the interview data were transcribed to text in electronic format for analytic purposes. All text was uploaded into the qualitative software package NVivo (QSR International) to assist the study team with the identification of codes related to recurrent treatment themes for both surgeons and patients. The interview data were evaluated for themes, and intercoder reliability was measured by comparing themes identified by study team members independently of one another. From there, we developed a list of important attributes and formatted the data displays and text with assistance from stakeholders to ensure that it was easily understandable by patients.
“Tailoring” refers to a process by which information about a person is used to shape the content and format of messages for that individual.15-17 A robust literature supports the efficacy of tailored messages across a range of audiences and health conditions.18-23 Expertise in message tailoring for this project was provided by the University of Michigan's Center for Health Communications Research (CHCR). CHCR has developed >150 tailored print-, telephone-, computer-, web-, and mobile-based interventions24-28 that have been carried out in settings such as health clinics, emergency departments (EDs), primary care offices, and schools. The tool provides a tabular presentation of the different aspects of several operations tailored to the patient's characteristics (eg, BMI, excess body weight, preferred characteristics of a surgical procedure) resulting in a permutation for each individual patient.
The conjoint analysis and the first version of our decision tool were completed and evaluated simultaneously in a survey fielded between May 2015 and January 2016. Development of the conjoint analysis and the decision tool involved parallel work streams with parallel results. The conjoint analysis was not intended to inform the tool but rather to provide more generalizable knowledge about patient preferences with bariatric surgery. All study participants were recruited from hospitals participating in the MBSC. The recruitment process is described in the “Participants (Target Sample Size = 1000 Patients)” section.
The MBSC is a regional, voluntary consortium of hospitals and surgeons performing bariatric surgery in Michigan. The goal of the MBSC is to improve the quality of care for patients undergoing bariatric surgery. To do this, the participating hospitals submit data to the MBSC clinical outcomes registry. Three times per year, the group meets to examine these data and to design and implement changes in care to improve the outcomes of care for bariatric patients. The project is funded by Blue Cross and Blue Shield of Michigan and coordinated by faculty and staff members from the Center for Healthcare Outcomes and Policy at the University of Michigan.
The MBSC was established in 2005 and is now composed of all 41 bariatric surgery programs in Michigan, along with an additional participating site in California. It enrolls approximately 7500 patients per year in its clinical registry, with nearly 100 000 patients currently in the registry. Participating hospitals submit data from a review of the medical records for all of their bariatric surgery patients. This review is conducted for each patient at 30 days after surgery. The information collected includes preoperative clinical characteristics and conditions, as well as perioperative clinical care and outcomes. The medical record reviews are performed by centrally trained, nurse data abstractors using a standardized and validated instrument. Each participating hospital undergoes a rigorous data accuracy audit annually to verify the accuracy and completeness of their MBSC clinical registry data.
In addition to the medical record reviews, the MBSC surveys patients who consent to participate at baseline and at 1, 2, 3, 4, and 5 years following their bariatric procedure. These patient surveys are the primary means by which data on long-term outcomes of care, including weight loss, QOL, comorbidity resolution, and patient satisfaction are obtained. Current rates of follow-up for these patient-reported outcome surveys based on year are listed in Table 5.
MBSC PRO Survey Follow-Up Rates by Year.
The 2 primary outcomes for the decision tool evaluation were decisional and clinical patient outcomes. The subcategories of each are described below. As secondary outcomes of the tool's usability and utility to patients, we used survey questions and data from our decision tool online infrastructure, such as the number of clicks or time spent on sections of the tool.
Patients in both the intervention and control groups completed (1) a baseline survey and (2) a follow-up survey 3 months after baseline survey completion (median number of follow-up days, 98 days for the control patients and 97 days for the intervention patients). Participants could access the decision tool from their personal device at any time.
“It was the right decision.”
“I regret the choice that was made.”
“I would go for the same choice if I had to do it over again.”
“The choice did me a lot of harm.”
“The decision was a wise one.”
Each item contains a statement regarding the decision being assessed for regret; 3 of the items are framed positively (eg, “It was the right decision”), and 2 items were framed negatively (eg, “The choice did me a lot of harm”). The 2 negatively framed items are reverse scored to ensure each item is scored in a consistent direction. Scores for each element of the scale are reported on a 0 to 100 scale, which is calculated by subtracting 1 from each item and then multiplying by 25. The overall score is also reported on a 0 to 100 scale by calculating the mean score on each of the 5 items. A score of 100 means no decisional regret, and a score of 0 means high regret.
Patients in both the intervention and control groups completed the usual MBSC baseline and follow-up surveys at 1 year. Measures obtained at these time points include weight, QOL, comorbidity resolution, and patient satisfaction.
Respondents were recruited from bariatric surgery information sessions around Michigan. A priori power calculations were not used to determine the target sample size due to time constraints and available resources. Information packets were distributed to attendees, and they could register to participate regardless of their immediate decision to consider surgery. To be eligible to participate, patients were required to have access to a computer/tablet/smartphone/other device to complete required electronic surveys at their convenience (as the tool and other surveys were all online), and they needed to meet the requirements listed below. These requirements were explicitly stated in the decision tool recruitment materials.
Respondents were given a $50 incentive. The University of Michigan IRB approved all study procedures associated with this component of the study.
We used the urn randomization method to assign participants to either the control or intervention group. The urn design is the most studied member of adaptive coin designs and is a compromise between designs that yield perfect balance in treatment assignments and complete randomization, which eliminates experimental bias.29
The following information was previously published in an article by Rozier et al.14 Copyright permissions for this information can be found in Appendix B.
The information is laid out as follows:
Stratified Analysis by Gender.
Stratified Analysis by Age.
Stratified Analysis by Income.
Latent Class Analysis.
Using a fractional factorial design,30 64 full profiles of bariatric surgery procedures were generated. Respondents were randomized to 1 of 8 question blocks. A fractional factorial design is a subset of a full factorial design, chosen to pair each attribute with and against other attributes in the study. The conjoint analysis was separate from the phase 1 study and was conducted in parallel to that analysis rather than informing the development of the decision aid.
Cognitive pre-tests were conducted to assure face validity of the survey instrument and comprehensibility of the survey method. We conducted iterative pre-tests on paper (n=7) and, after subsequent revisions, conducted additional pre-tests in an online format (n=10). Respondents were provided detailed pre-test questions and were given the option to verbally share comments during the pre-test.
The survey instrument contained an introductory section that explained the task and defined the attributes and levels used in the survey. Respondents then answered one practice discrete choice question that was not scored and 8 scored discrete choice questions. Discrete choice questions presented 2 hypothetical bariatric surgery profiles. Respondents were asked to select which procedure they would prefer or indicate that they would not choose either procedure (Figure 7). An opt-out option was important to include in the experimental design because patients who initially would consider a surgical option for weight loss may ultimately decide against any available option based on combination of attributes in the profiles. Finally, respondents provided demographic information and indicated if they found the discrete choice questions difficult to answer and, if so, why they were difficult.
Sample Discrete Choice Task.
We estimated the relative value of harm and benefits for leading weight-loss surgical options. We calculated marginal willingness to pay for procedure attributes by asking patients to choose a dollar amount (ie: $100, $500, $1000, $2000, $5000, $10,000 & $15,000) that they would be willing to pay out of pocket for a particular treatment and its associated risk of harm and benefit. We conducted a latent class analysis to identify respondent subgroups.
The primary analysis used an effects coded conditional logit adjusted for clustering at the respondent level.10,31 Stratified analyses were conducted based on gender (male/female), income level (≤$25,000/>$25,000), and age (<45 years/≥45 years). A secondary analysis was restricted to the 90% of respondents who indicated they did not find the conjoint questions difficult to answer. All regression analyses were conducted using Stata 13.
An additional analysis used a model where initial out-of-pocket cost, risk of complications, total weight loss, and years treatment has been available were all recoded as continuous variables. Recoding initial out-of-pocket costs as a continuous variable also allowed us to calculate marginal willingness to pay for surgery.
Additional analyses recoded certain variables as dichotomous instead of the effects coding (see Table 4 for the attributes and their levels). The extensiveness attribute was separated into two separate variables: one based on weeks of recovery and one based on reversibility. We combined years of treatment into two levels instead of four: newer (1 and 5 years); established (10 and 20 years). We separated weight loss into two separate variables based on amount lost during year 1 and amount regained during year 2. We combined risk of complication into two levels instead of four: low complication rate (0% and 5%); high complication rate (20% and 35%).
Finally, we conducted a latent class analysis to determine whether there were unobserved subgroups whose preferences were driven by particular attributes more than other respondents.32 The cluster model of latent class analysis is conducted by introducing dummy variables into the conditional logit model based on the number of classes that may plausibly exist. A maximum likelihood estimation then determines the probability that a respondent would provide the observed responses, and the model that best categorizes the respondents is used. The latent class analysis was conducted using Latent Gold (Version 5.1; Statistical Innovations, Inc.).
A total of 815/878 (93%) people completed the discrete choice experiment (Table 4), making this the largest study on patient preferences for bariatric surgery to date. Respondents were much more likely to be female (79.9%) than male (20.1%). Nearly 40% were aged 30-59 years. The distribution of gender and age is typical of the population seeking bariatric surgery.33 The racial and ethnic backgrounds reflect Michigan's general population, with the majority identifying as white (75.2%), and a large minority identifying as black or African-American (17.1%). A small number of respondents identified as other racial or ethnic minorities. Other characteristics such as marital status, income, and education were similar to the general population of the state. As expected, respondents were more likely to indicate having a medical condition (27.4% with diabetes; 47.4% with arthritis) or be disabled (15.7%) than the general population (Table 6).
Characteristics of Conjoint Respondents.
A coefficient value from a discrete choice experiment that is greater than 0 indicates that including that attribute level in a surgical profile makes it more likely that the profile will be selected. The range of values for coefficient is −1 to 1. For example, the coefficient for “loss of 80% of excess body weight” is .460. This means a surgical profile (i.e., the profile of a surgical procedure) was more likely to be selected if it included that attribute level. A coefficient value less than 0 indicates a profile is less likely to be chosen when it included that attribute level. For example, the coefficient for “loss of 20% of excess body weight” is −.473. This means a surgical profile was less likely to be selected if it included that attribute level of weight loss of only 20%.
The most significant attributes that drove respondent choice were resolution of medical conditions (e.g. improvement of diabetes control), total weight loss, and initial out of pocket costs (Figure 8; Table 7). A profile consisting of resolution of sleep apnea, hypertension, high cholesterol, and diabetes was more likely to be chosen by a participant, while a profile consisting of failure to resolve any of those conditions made it less likely to be selected. Partial resolution of medical conditions did not have a significant influence positively or negatively on a patient's choice of surgery. Each amount of additional weight loss made a surgical profile more likely to be selected, with a loss of 80% of excess body weight as the most significant positive characteristic for the entire study. Each amount of additional cost made a surgical profile less likely to be selected, with an initial out-of-pocket expense of $15,000 as the most significant negative characteristic in the study.
Attribute Parameters from Discrete Choice Experiment.
Attribute Parameters and Marginal Willingness to Pay.
Other attributes had less influence on respondent choice than the three described above. Respondent choice was not greatly influenced by how treatment works, the number of years a treatment has been available, and the required dietary changes, although certain levels of each of these attributes had positive effects on respondent choice. For example, a profile that had been available for 20 years, had 0% or 5% risk of complications, had no side effects, or had no dietary restrictions was more likely to be selected. However, these and many other factors had minimal influence on patient preference when compared to total weight loss, medical conditions, and costs.
Males were more likely to accept a treatment that restricts food and decreases hunger. Females were slightly more tolerant of risk than males, although the difference was not statistically significant (Figure 9). When compared to older respondents, younger respondents (18-44 years) were more likely to select procedures with greater weight loss, procedures with low out-of-pocket costs, and procedures that reduce absorption (Figure 10). Although it was not a statistically significant difference, low-income respondents were more sensitive to initial out-of-pocket costs. Otherwise, income level (greater or less than $25,000 per year) did not significantly affect responses (Figure 11).
Respondents were willing to pay more for procedures that resulted in the greater weight loss ($5470 for every 20% of total weight loss) and that resolved weight-related medical conditions ($12,843). They were also willing to pay higher amounts for a non-reversible procedure with a 2-week recovery time ($10,534) and a procedure with no side effects ($7,808). Respondents showed the least willingness to pay for attributes such as dietary changes, number of year's treatment has been available, and how treatment works.
Latent class analysis indicated three unobserved groups of respondents (Figure 12). The first class, “benefit-focused,” was the largest, composed of 38.3% of all respondents, and was primarily concerned with weight loss and the resolution of weight-related medical concerns. The second class, “cost-sensitive,” was most concerned with out-of-pocket costs and was made up by 31.4% of all respondents. The decisions of those in the third class, “procedure-focused,” 30.3% of all respondents, were balanced across all attributes, with slightly greater concern given to how the treatment itself worked. When matched with demographic characteristics, none of the three groups were more likely than another to have a particular identity (gender, income, or age).
This portion of our study used a novel approach, conjoint analysis, to evaluate the attributes of the four most common surgical procedures for weight loss in the United States – RYGB, adjustable gastric banding, sleeve gastrectomy, and duodenal switch.34 We found that excess weight loss, initial out-of-pocket costs, and resolution of medical conditions are most important in patient decision-making. These preferences differed little by key demographic characteristics, although low-income respondents were more sensitive to costs, younger respondents were more sensitive to excess weight loss, and males were more sensitive to resolution of medical conditions. A marginal willingness-to-pay analysis showed that respondents were willing to spend several thousand dollars for the most important characteristics, such as high excess weight loss and resolution of medical conditions.
Our intervention consists of an interactive, web-based decision support tool that provides patients and their physicians with a report of customized, patient-specific estimates of the risks and benefits of the treatment alternatives. See Figure 13 for a screenshot of the phase 1 tool.
Screenshot of Phase 1 Tool (Treatment Table).
Patients in the intervention group electronically completed the MBSC baseline survey (demographics and baseline health status), conjoint analysis, and the decision support tool, including customized estimates of the risks and benefits of the treatment options based on their specific body weight and comorbidity status. They also received a detailed summary of their information to use when discussing treatment options with their physician. With patient exposure data collected online, we were able to obtain some information regarding the length of time patients had the decision tool website open, but there was wide variation, ranging from about 2 minutes to roughly 3 months, making it difficult to know how much time patients actually spent viewing the decision tool materials vs simply having the website window open on their device; this suggests that these data are unreliable for estimating the level of exposure to the intervention.
Patients in the control group completed the MBSC baseline survey (demographics and baseline health status) and conjoint analysis. They did not complete any aspect of the decision support tool.
The 2 primary outcomes for the decision tool evaluation were decisional and clinical patient outcomes. The subcategories of each are the same as for the phase 1 tool and are described in the “Methods” section for phase 1.
For discrete-choice experiments, estimates of sample size followed from the design of the choice set to maximize the efficiency of the experiment.9,10 For this survey, we used a fractional factorial design to identify specific subsets of paired choices to maximize the statistical power needed to estimate key main effects.
For the mixed models and generalized linear mixed models, there are no closed-form equations for calculating sample size. We conducted a Monte Carlo simulation analysis varying the distributions of the surgeon-specific variable and the decision tool exposure to develop estimates of our power to detect significant differences in our primary outcome of interest, the Health and Activities Limitation Index (HALex, a continuous measure of QOL). In our power analyses, we explicitly accounted for the clustering of patients within surgeons36,37 by varying the intracluster correlation coefficient (ICC) between 0.01 and 0.2. These values reflect the range of ICCs observed in other large studies with patient clustering.38 Based on preliminary estimates, we assumed the SD of HALex to be 0.18. Using this, we determined that with 60 surgeons and 45 procedures per surgeon per year (conservative estimates based on historical data of consented patients treated in 2010 who returned 1-year follow-up surveys), we would have 80% power using a 2-sided test at α = .05 to detect mean differences of between 0.017 (best-case scenario) and 0.047 (worst-case scenario) as statistically significant.
We aimed to compare the decision tool with usual care to determine its effects on decisional outcome measures and patient outcome measures at baseline and at the 3-month and 1-year follow-up periods. Participants were randomly assigned to either the intervention group, which was exposed to the decision tool (n = 440), or to the control group (n = 438), which did not complete the decision tool.
The data for each variable were examined to identify potential outliers, extreme values, missing data, and possible coding errors. All consented participants who returned surveys at baseline and at each follow-up period were included in the analysis. We performed a complete-case analysis in which participants with missing surveys were excluded from the analyses. We did not compare the descriptors of those who did and did not return the surveys. Efforts were made to clean the data, and, where possible, queries about problematic data were directed to the bariatric surgery program to ensure that the data were as accurate as possible.
We used linear mixed models with a normal link function for continuous outcomes and generalized linear mixed models with a logit link function for binary outcomes. The outcomes were modeled as a function of the intervention group, adjusting for procedure-specific variables and patient-specific clinical and demographic variables. We assumed fixed coefficients for the patient- and procedure-specific covariates.
Models were developed according to the following steps. We examined univariate models of each outcome and patient and procedure factors to determine which factors were predictive of the outcomes (P < .05). We then included all factors that were predictive in the univariate models in a multivariable model and excluded any factors with P > .20 from the multivariable model. Analyses were completed using SAS 9.4 (SAS Institute).
We randomly assigned a total of 878 participants for this study. Of those, 438 participants were randomly assigned to the intervention group, and 440 participants were randomly assigned to the control group. In the intervention group, 35% (n = 154) of the participants randomly assigned to the phase 1 decision tool underwent bariatric surgery, and 38% (n = 166) of participants randomly assigned to the control group underwent bariatric surgery. All 154 participants from the intervention group who had surgery completed the 3-month follow-up survey, and 155 participants from the control group who had surgery completed their 3-month follow-up survey; 11 participants did not respond to questions on decisional regret (Figure 14).
Participant Flow Diagram.
Randomization yielded balanced patient populations. Univariate comparisons between the intervention group (n = 438) and the control group (n = 440) only showed a statistically significant difference in the proportion of diabetic participants (23.5% in the intervention group vs 30.0% in the control group, P < .05) (Table 8a).
Comparison of Patient Populations: Phase 1 Decision Tool Intervention Group vs Control Group (All Patients).
Among participants who followed through and had surgery from the intervention group (n = 154) and the control group (n = 166), standardized differences with magnitudes greater than >0.10 (ie, >0.10 and <−0.10) reveal imbalances in the following characteristics: gender, race, income, education, hypertension, and hyperlipidemia (Table 8b).
Baseline Characteristics: Phase 1 Decision Tool Intervention Group vs Control Group (Surgical Patients).
Multivariate logistic regression (for categorical outcomes) and linear regression (for continuous outcomes) yielded no statistically significant differences between the intervention group and the control group for 30-day outcomes/complications, 1-year weight change, satisfaction, and comorbidity resolution (see Table 9, which also shows additional information regarding the mean differences for each measure).
Comparison of Clinical Outcomes: Phase 1 Decision Tool Intervention Group vs Control Group (Surgical Patients).
An evaluation of measures of decisional regret between the intervention group and the control group did not reveal any statistically significant differences (Table 10). The summary decisional regret scores for the intervention and control groups were 80.9 and 83.8, respectively (P = .20). Individual measures of decisional regret also did not vary significantly.
Comparison of Decisional Regret: Phase 1 Decision Tool Intervention Group vs Control Group (Surgical Patients).
We assessed patient interaction with the decision tool through survey questions and data from our decision tool online infrastructure. In the tool, a main data table presented most of the customized information for patients. In all, 100% of the 154 participants randomly assigned to the intervention who had surgery viewed the table given the sequence of the decision tool's data presentation, but very few interacted with the important, personalized information contained in the table. We found that only 63% of 154 participants explored the data table. For example, only 31.4% of the 154 participants viewed any of the procedural videos that provided details about the conduct of each operation. Also, only 15% of 154 participants expanded the table fields about the benefits of surgery, while even fewer (8.3%) explored the specific risks portion of the table.
This portion of the study presented patients with a decision support tool with highly customized estimates of the risks and benefits of various treatment options. Although the tool was favorably reviewed during usability testing (as described in aim 1), once it was incorporated into the perioperative workflow of bariatric surgery practices in our study population, it did not demonstrate significant changes in treatment choice, knowledge, treatment-preference concordance, or decisional conflict. There were also no significant differences between the intervention and the control cohorts. Finally, once the tool was implemented, usability statistics showed that study patients seldom accessed the tool's individualized content (eg, clinical options table and personalized estimates of harm and benefit) after initial enrollment.
The third component of our study involved the revision of the first iteration of our decision support tool to better meet the needs of our patient population. Although we feel that the decision tool we created is thorough, we discovered via discussions with the PCORI clinical effectiveness research program that there was some concern about the possibly excessive length of the final product that could adversely impact usability and patient engagement with key components. Therefore, we halted recruitment of participants. There were 881 participants recruited at this time (targeted recruitment was 1000 participants). We did not have full long-term follow-up data at the time the decision was made to revise the tool. The utilization data informed our understanding of how participants used the tool and their impression of the tool's value. The initial tool has 3 main components: (1) the baseline survey, (2) the conjoint analysis, and the (3) decision aid. There was a >98% completion rate of each component; however, there was little user engagement with the content provided. For example, only 4% to 17% of users clicked on the information screens for the various treatment options. This was especially concerning, as these screens are the key components of the intervention that present all the pros and cons of the different surgical procedures. Despite these objective metrics, nearly 50% of participants felt that the decision aid was personalized for them, indicating a clear opportunity to improve the final patient tool. Therefore, we were granted an 18-month no-cost extension to facilitate revision and evaluation of an updated tool.
To gather feedback and suggestions for improvement, we conducted 27 one-on-one interviews with participants who completed the first iteration of the tool. These interviews were audio-recorded and transcribed for qualitative analyses. Several themes emerged as a result of these interviews, including the following:
The decision tool was revised based on the aforementioned feedback. The same risk and benefit models were used as in the first iteration of the tool. Major changes included the following:
We aimed to compare the revised decision tool with usual care to determine its effects on decisional outcome measures and patient outcome measures at baseline and at the 3-month follow-up period. Patients were recruited to participate in the intervention group (n = 311). In this phase of the study, we did not randomly assign participants due to the condensed time frame and because aim 3 was an extension of the original work plan. For the control group, we selected a propensity-matched cohort of participants from the phase 1 control group. We could not match 1 intervention group participant to a control; therefore, our final analytic population included 310 participants. We also compared patient usability statistics between the phase 1 and phase 2 decision tools. Data cleaning and analyses were performed using the same procedures as in the analyses involving the first iteration of the decision tool.
Respondents were recruited from bariatric surgery preoperative appointments from various programs around Michigan. Our target sample size was 300 participants. The recruitment criteria were identical to those described previously for the first phase of the project. The only difference was changing the timing of recruitment to the preoperative appointment to maximize the number of participants who would progress to surgery. The University of Michigan IRB approved all study procedures associated with this component of the study as well.
Participants were not randomly assigned for the assessment of the revised iteration of the tool. Rather, all respondents who completed the MBSC baseline survey (demographics and baseline health status) received the decision support tool and the customized estimates of the risks and benefits of the treatment options based on their specific weight and comorbidity status. As in the aim 2 RCT, they also received a detailed summary of their individualized predicted harms and benefits of the surgical options to use when discussing treatment options with their physician, as well as periodic “check-ins” for 6 months following baseline completion.
For the control group, we assembled a cohort from the phase 2 RCT control participants that matched our phase 2 intervention participants using propensity matching to balance the groups based on baseline patient characteristics (gender, race, income, education), comorbidities (hypertension, hyperlipidemia, diabetes, psychiatric disorder, arthritis, gastroesophageal reflux disease [GERD], asthma, sleep apnea, heart disease, and current smoking status), and surgery performed (RYGB, sleeve gastrectomy, duodenal switch). Patient characteristics that were collected at phase 2 baseline were considered candidate characteristics for matching, and we narrowed the field of matching criteria to include characteristics associated with use of the decision tool at the P < .20 level. This resulted in 1:1 matched control and intervention groups that each included 310 patients. With caliper set at 0.20, when >1 control participant was an eligible match with an intervention participant, we selected the participant with the best match. Among these matched participants, the only noteworthy standardized differences (difference >0.10 and <−0.10) were in age (−0.1705), baseline BMI (0.1468), and income (−0.1123). Thus, we created a data set that consisted of 310 participants who were exposed to the revised tool and 310 control participants who tested the first iteration of the decision tool, and we used multivariate modeling to compare the study outcomes between these 2 groups.
The matched participants included 154 phase 2 intervention participants and 159 of the matched control participants, all of whom underwent a bariatric surgical procedure. Exposure to the decision tool was carried out in the same way as in the phase 1 RCT. Notable standardized differences between the groups who had surgery were observed in age (−0.1467), baseline BMI (0.1424), gender (0.1603), income (0.2236), education (0.2831), and type of surgery performed (0.3943). Both groups were patients who underwent surgery during the evaluation period (July 2017 to September 2018).
We also compared participants using the first iteration of the tool with those using the second iteration.
These outcomes were the same as those described in detail in the aim 2 methods section. We examined differences in patient characteristics and clinical outcomes between the original control group and the phase 2 cohort. We then compared measures of decisional regret and tool utility between the phase 2 and the phase 1 cohorts (see the description of the decisional regret scale for aim 2).
The second iteration of the tool was fielded between July 2017 and September 2018.
We conducted 2 online surveys, with 1 survey at baseline and a follow-up survey 3 months after the baseline survey completion (median follow-up survey completion time, 94 days postbaseline). Candidates were recruited from the preoperative surgical clinic in practices throughout Michigan.
Models were developed according to the following steps. We examined univariate models of each outcome and patient and procedure factors (see Table 9) to determine which factors were predictive of the outcomes (P < .05). We then included all factors that were predictive in the univariate models in a multivariable model and excluded any factors with P > .20 in the multivariable model. The models were reviewed by clinicians to evaluate clinical plausibility and validity. Analyses were completed using SAS 9.4.
About half of the participants in each cohort followed through with surgery within the study period (n = 159 control participants, n = 154 intervention participants). Among the participants who had surgery, more control participants opted for sleeve gastrectomy (88.1% vs 75.3% for intervention participants), and more intervention participants opted for RYGB (24.7% vs 10.7% for control participants) (Tables 11a and 11b).
Comparison of Patient Populations: Phase 2 Decision Tool Intervention Group vs Propensity-Matched Phase 1 Control Group (Participants Who Had Surgery and Participants Who Did Not Have Surgery).
Comparison of Patient Populations: Phase 2 Decision Tool Intervention Group vs Propensity-Matched Phase 1 Control Group (All Patients Had Surgery).
We then tested for differences between the intervention group and the propensity-matched phase 1 control group in 30-day outcomes among the subset of participants who had surgery. We first examined differences in decisional regret between these 2 groups (Table 12). Multivariate regression analyses determined that there were statistically significant differences between the phase 2 intervention participants and the propensity-matched phase 1 control participants. Phase 2 intervention participants scored higher (ie, less decisional regret) in overall decisional regret (P = .0118) and when answering the questions, “I made the right decision” (P = .0452), “The choice did me a lot of harm” (P = .0106), and “The decision was a wise one” (P = .0095).
Comparison of Decisional Regret: Phase 2 Decision Tool Intervention Group vs Propensity-Matched Phase 1 Control Group.
Using multivariate logistic regression to control for patient factors and comorbidities, we found a significant difference in the ED visit rates between the 2 cohorts; control participants (12.0%) had a higher ED visit rate than that of intervention participants (5.2%) (P = .0242). There was also a difference in participants with at least 1 utilization event within 30 days, in which control participants (14.5%) were more likely than intervention participants (8.4%) to have a utilization event (P = .0301). All other 30-day outcomes showed no significant difference between the control and intervention groups (P > .05 for each) (Table 13).
Comparison of Clinical Outcomes: Phase 2 Decision Tool Intervention Group vs Propensity-Matched Phase 1 Control Group.
We compared characteristics, decisional regret, and usability between the phase 1 and 2 decision tools (see Tables 14-17). Please refer to Figure 14 for a description of where some participants were lost to follow-up. It is crucial to note that although 155 participants from phase 2 underwent surgery within the study time frame, 161 participants completed the follow-up survey assessing decisional regret and decision tool utility. The additional 6 respondents were presumably referring to their decision NOT to have surgery yet or their decision regarding surgery in the future. Overall, we found that participants reported the revised phase 2 decision tool as being more useful than the phase 1 version. Those who used the phase 2 tool found the tailored benefits and risks portion to be more helpful than did those using the phase 1 tool (48% vs 26%, respectively; P < .001). There was also longer and more in-depth interaction with the phase 2 tool as measured by time using it. With the phase 2 tool, 43% of participants reported spending more than an hour with the tool compared with only 22% of participants with the phase 1 tool (P < .001). Additionally, participants interacted with the phase 2 tool significantly more than with the phase 1 tool, with more clicks on the various aspects of the tailored information presented. Finally, there were no differences in overall decisional regret between participants who used the tool in phase 1 vs the tool in phase 2 (P = .9875).
Comparison of Patient Populations: Phase 1 Decision Tool vs Phase 2 Decision Tool.
Comparison of Patient Populations: Phase 1 Decision Tool vs Phase 2 Decision Tool (Surgery Patients Only).
Comparison of Decisional Regret: Phase 1 Decision Tool vs Phase 2 Decision Tool.
Comparison of Decision Aid Utility: Phase 1 Decision Tool vs Phase 2 Decision Tool.
Multivariate regression models incorporated baseline patient factors that were statistically indicated as covariates with P < .20 in the univariate comparisons of the intervention and MBSC registry groups (see Table 17). Multivariate logistic regression (for categorical outcomes) and linear regression (for continuous outcomes) found that the phase 1 intervention group was significantly less likely to experience any surgical complications than the phase 2 group (3.7% vs 10.3%, respectively; P = .0175) but showed no significant difference in the utilization of additional health care resources (10.4% vs 8.4%, P = .3674), specifically, ED visits (8.8% vs 5.2%, P = .1408). We were not able to make meaningful comparisons between other specific outcomes (ie, specific complications) due to low event rates. As such, statistical measures of difference were not included in the table.
Thirty-Day Outcomes: Phase 1 Intervention vs Phase 2 Intervention.
Our study had several key findings. First, through our novel application of a conjoint analysis, we found that excess weight loss, initial out-of-pocket costs, and resolution of medical conditions are most important in patient decision-making when choosing a bariatric surgical procedure. These preferences differed little by key demographic characteristics, although respondents with low income were more sensitive to costs, younger respondents were more sensitive to excess weight loss, and men were more sensitive to the resolution of medical conditions. Second, we found that the first iteration of the decision tool did not result in clinically meaningful reductions in decisional regret, improvements in surgical satisfaction, or better clinical outcomes compared with usual care. However, given the low number of participants who were randomly assigned to the decision tool and actually viewed the data about the risks and benefits, this experiment is perhaps best understood as having inconclusive findings. Third, we learned that given the volume of information available, patients require a more condensed decision tool that is simpler to use. Therefore, we revised and tested the decision tool in a post hoc exploratory analysis using historical controls. In our testing, we found that compared with propensity-matched controls, a streamlined, personalized risk assessment tool that included testimonial information from other patients was associated with improvements in decisional regret scores among patients who had surgery. Additionally, health care utilization, such as ED visits, was lower in phase 2 with the revised decision tool. These findings are strong signals, but they must be confirmed in an RCT before we can conclude that the decision tool improves decision-making.
Aside from the clinical risks and benefits of the different bariatric options, there are many other attributes of the procedures that should be considered by patients. Our conjoint analysis sought to better understand which attributes our population values in making a decision and to what degree. There has been some previous work evaluating these preferences. One prospective study asked patients the primary reasons for selecting a particular procedure and found different motivations for each: safety and less invasiveness for adjustable gastric banding; the rigid dietary requirement to avoid dumping syndrome for RYGB; and the amount of weight loss for duodenal switch.39 A retrospective study found similar variability in reasoning: RYGB patients cited the evidence base and success rate, sleeve gastrectomy patients referenced provider recommendations, and adjustable gastric banding patients cited the reversibility and less invasive nature.40 Yet another study found that most patients had made a decision about which procedure to pursue before consultation with a surgeon and that maximum weight loss was the respondents' primary concern.41 One study specifically evaluated the value patients place on weight loss as a deciding factor and found that most patients pursued surgery for health reasons and also had unrealistic expectations of weight loss.42 Our latent class analysis aligns with the results of previous studies that ask patients why they choose to avoid certain procedures.40-42 These studies have found subgroups of patients who are motivated by ≥1 aspect of a given procedure. Our results show 3 subgroups: those motivated primarily by (1) cost, (2) excess weight loss and resolution of medical conditions, and (3) reversibility of and recovery from the procedure. As in previous studies, the subgroups were not easily divided along demographic lines with random distributions of various demographic variables across the 3 subgroups.
One unique aspect of our study is the importance of initial out-of-pocket costs as a driver of patient preferences. Among high-income countries, the influence of cost may be unique to the United States. Most studies from countries with more comprehensive coverage have not taken this into account when measuring patients' choices for bariatric surgery. The findings from this study suggest that providers in the United States should strongly take into account the health insurance benefits of their patients.
Currently, the 2 most commonly performed bariatric surgical operations in the United States are sleeve gastrectomy and RYGB. Surgeons are often engaged in conversations with their patients about the risks and benefits of each, and patients often present with their own preference for an operation. Patient satisfaction after bariatric surgery is of paramount importance, as this important decision can lead to significant physical and psychosocial changes.33 During our study period, sleeve gastrectomy overtook RYGB as the predominant primary bariatric operation. This may influence the available information and education provided to patients. Moreover, this change in popularity is likely a result of surgeon preference for an operation that requires less technical skill and has fewer long-term complications. Our study's results would not have predicted the rise in popularity of sleeve gastrectomy compared with RYGB.43,44 Specifically, patients tend to value more weight loss, which occurs with RYGB and not sleeve gastrectomy. Therefore, the most likely explanation for the trend toward more sleeve gastrectomy stems from provider recommendation, which is not accounted for in this study, rather than patient preference. In the context of shared decision-making, there are serious challenges in designing evaluations of decision tools for bariatric surgery that account for the shifts in practice patterns in the field (ie, movement toward recommending one procedure over another).
Advances in the field of shared decision-making have resulted in greater patient involvement in health decisions in general. However, low health literacy poses major challenges to shared decision-making, and because morbid obesity disproportionately affects ethnically and racially diverse people and those of low socioeconomic status, it was important that our program was specifically tailored to address the barriers to communication of health information in such populations. Shared decision-making requires that patients understand the nature of their disease or condition, understand the risks and benefits of the available treatment options, have considered their own preferences, have participated in decision-making to the extent they wish, and have made a decision that is consistent with their preferences and values. There are dynamics in the physician-patient relationship that are difficult to overcome even with the best available patient education. Surgeons bring their own preferences and biases into discussions about surgery. These differences must be reconciled in the clinic and as best as possible reflect the patient's values and goals. Nonetheless, we believe that the long-term results of surgery may be positively impacted with improved patient education and expectations in the pre- and perioperative periods. Interactive, computer-based decision aids have been shown to be more engaging to patients and have been used successfully with patients who have little or no experience using computers as well as in those with low literacy.45 Our work used the best available research on engaging patients with limited literacy (including language, presentation of data on risks and benefits, visual displays, and format) to guide the development of our decision tool.46
Bariatric surgery, like spine surgery and breast and prostate cancer treatment, is considered a highly preference-sensitive medical issue. Existing decision aids for bariatric surgery have many limitations.8 First, they are frequently out of date. For example, the Foundation for Medical Decision Making's video-based decision aid for bariatric surgery does not include any information about sleeve gastrectomy. This decision aid does not provide up-to-date information about the risks of complications and mortality with bariatric surgery, citing risks that are far higher than the best current data indicate. Existing decision aids in bariatric surgery are also limited in that they provide information about the average comparative risks and benefits of the treatment options but do not provide estimates of the risks and benefits of the different procedures that are tailored to the characteristics of individual patients. As a result of these drawbacks, decision aids are not frequently used in making treatment decisions in bariatric surgery. Our decision tool is highly innovative in that it integrates data from a large clinical registry with individual patient data to provide patients with real-time, customized, accurate information regarding the risks and benefits of the treatment options to better inform decision-making. Also, our tool will be continuously updated to ensure that the data it provides on risks and benefits are accurate and current. Our tool will also provide information about other attributes of the treatment options that bariatric surgery patients and other relevant stakeholders feel are important to consider in deciding whether to have bariatric surgery and which type.
We were able to demonstrate statistically significant improvements in patient decisional regret using the revised decision tool in the intervention group vs the control group. However, patient decisional regret did not improve in a comparison of the revised decision tool with the original tool. Nonetheless, with the positive qualitative feedback received from patients, surgeons, and allied health professionals, we are moving forward with broader dissemination. As part of the project, this intervention will be disseminated throughout the state of Michigan. We will also make it available for dissemination nationally via cooperative relationships between the MBSC and the American Society for Metabolic and Bariatric Surgery and the American College of Surgeons. Also, leveraging our invaluable academic environment, we have begun collaborating with the University of Michigan's Fast Forward Medical Innovation (FFMI) program to develop a successful business case to find funding and develop strategic partnerships to ensure the longevity and wider dissemination of our decision tool. As part of this, members of our team are participating in the FFMI fastPace Biomedical and Innovation Course. The goal of this course is to develop a successful business case used to secure funding and attract collaborators and to expand our network of innovation partners, mentors, and potential stakeholders.
There are several potential areas for future research. First, we could glean additional information from an RCT of the phase 2 decision tool with improved follow-up rates as a priority. Second, we continue to work toward translation of the tool into other languages, particularly Spanish. Third, we plan to validate the MBSC Registry risk prediction tool. The goal with future validation would be to demonstrate the level of agreement between predicted probabilities of benefits and harms from the surgical models. We will use the established Transparent Reporting of multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD) reporting guidelines47 to allow for decision makers to appropriately assess the accuracy of the predictions.
We were very sensitive to the need to include key characteristics of subpopulations, particularly the most vulnerable. We included important covariates into our modeling to determine the weights of factors external to the choice attributes relative to the weight of the choice attributes themselves. In particular, variation in preferences by subpopulations defined by race/ethnicity or by other patient characteristics identified in focus groups or from previous research were tested and incorporated into the decision tool. We did not perform subgroup analyses in the RCT portion of the study given the small sample sizes.
The most important limitation to our study conclusions is that we did not have the time or resources to complete a second RCT for the revised decision tool, which limits the internal and external validity of our conclusions regarding the revised tool. Second, the aim 1, 2, and 3 samples were convenience samples of patients at information sessions and preoperative visits. We do not have information on those who declined to participate in the survey. Additionally, respondents completed surveys after participating in information sessions. Therefore, if sessions were biased toward certain risks or benefits, respondent choice could be biased in similar ways. Third, we conducted this work in a single state, which may threaten the generalizability of our findings and of the decision tool. However, have previously conducted epidemiologic studies about bariatric surgery in Michigan and do not believe that the patients and providers in Michigan are systematically different from those in the rest of the country. Fourth, many of the attributes in our conjoint analysis had to be simplified. For example, we combined total weight loss and weight regain. Even after simplifying the list, we had a high number of attributes, given that many conjoint analyses will restrict to ≤5 attributes. Therefore, the complexity of the choice question may have impacted our results. Fifth, we were unable to include many important nonsurgical treatments, such as medications and meal replacements, which would make a worthwhile extension of this study. Sixth, the attributes and procedures we studied reflect bariatric surgery at a particular moment in time. The study began in 2015, and since that time, there has been some evolution in bariatric surgery practices. For example, sleeve gastrectomy now represents >70% of the primary bariatric operations in Michigan, and endoscopic balloons have become an uncommon but intriguing option for some patients. The underlying science behind bariatric surgery would no longer be described simply as limiting nutrient absorption vs restricting calories. As such, we now have evolving endoluminal therapies that continue to develop and gain some popularity. Seventh, we conducted a per-protocol analysis among the subgroup who underwent bariatric surgery. Therefore, while it appears there was a 65% loss to follow-up in the phase 1 decision tool trial, patients who did not have surgery were excluded from our outcome analyses since we were unable to distinguish between patients who had chosen a surgery and were still waiting to have it and those who actively chose no surgery; thus, we could not make any authoritative statements regarding their responses to the primary outcome questions about their surgical decision (ie, regret and satisfaction). Eighth, although the risk and benefit prediction models derived from the MBSC and used in our decision aid have been internally validated with patient populations in the MBSC, the levels of agreement between predicted and observed probabilities of benefits and harms from the surgical models have not been published and are not included in the report. External validation of the surgical models is underway. This will be an important topic for future research. Ninth, the preoperative visit served as the time of recruitment for the revised decision tool. As such, recruiting from the preoperative visit meant patients were more likely to undergo surgery. In contrast, participants in the RCT were recruited from “bariatric surgery information sessions,” which were presumably earlier (ie, time before surgery) than a preoperative visit. This means that comparisons between the trial participants and the cohort participants may be confounded by the time before surgery and where in the natural history of their decision-making the patient first saw the decision tool. As the length of time between exposure to the decision tool and surgery increases, any association between the effect of the decision tool and the decision made may be diluted. Matching or statistical adjustment cannot remove that source of confounding because there is no variation for the revised decision tool; all patients were recruited from a preoperative visit closer to the time of surgery and perhaps past the decision for the type of procedure. Tenth, we did not assess the rigor or attributes of the shared-decision-making discussion between the surgeon and the patient; therefore, we do not know what effect, if any, this may have had on the primary outcomes of our study. Finally, aim 3 was not part of the original study protocol: It was added after the results from the RCT were obtained. Post hoc analyses, especially those with historical controls, are always subject to validity concerns and are taken as a point of departure for confirmatory research rather than as established fact.
The decision to choose one weight loss method over another is complex and relies on factors related to the procedures, patient and surgeon preferences, and how providers frame choices. We often know why patients select one procedure over another, but the hypothetical trade-offs in this study give us the relative importance of these attributes and the potential effects on short- and long-term outcomes. Recognizing that patients will vary in how they understand these trade-offs can improve the patient-provider discussion on the risks and benefits of various weight loss options. This study provides information on which characteristics are most important from a patient perspective and shows that patients demonstrate heterogeneity in preferences. It also provides a decision tool with patient-specific information about the risks and benefits of the surgical options under consideration. Our ultimate goal is to provide patients with the necessary information and guidance to make a final decision that is congruent with their personal values and expectations. We were able to identify the relative importance of several risks and benefits of different weight loss approaches. We were limited in the RCT of our original decision tool, as not all patients pursued surgery, and despite attempts to recruit patients just before surgery, we did not find statistically significant changes in some of our primary outcomes, such as satisfaction and clinical outcomes. In the RCT, we found that our revised decision tool was qualitatively well received by patients and, in a post hoc analysis with historical controls, was associated with a reduction in decisional regret. We are encouraged by the positive feedback from patients and providers about the comprehensive nature of the tool and the potential for it to enable meaningful discussions about weight loss in this dyad.
Research reported in this report was funded through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (#CE-1304-6596). Further information available at: https://www.pcori.org/research-results/2013/testing-decision-aid-help-patients-choose-between-two-types-bariatric-surgery
PCORI Discussion Guide: March 25-29, 2014 (PDF, 101K)
JAMA Surgery Copyright Permission – Text and Tables/Figures (PDF, 3.1M)
Phase 1 and Phase 2 Tool vs MBSC Registry (PDF, 151K)
Ghaferi AA, Hawley ST, Fagerlin A, et al. (2021). Testing a Decision Aid to Help Patients Choose Between Two Types of Bariatric Surgery. Patient-Centered Outcomes Research Institute (PCORI). https://doi.org/10.25302/01.2021.CE.13046596
The [views, statements, opinions] presented in this report are solely the responsibility of the author(s) and do not necessarily represent the views of the Patient-Centered Outcomes Research Institute® (PCORI®), its Board of Governors or Methodology Committee.
This book is distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License which permits noncommercial use and distribution provided the original author(s) and source are credited. (See https://creativecommons.org/licenses/by-nc-nd/4.0/
Your browsing activity is empty.
Activity recording is turned off.
See more...
