Recent drug approvals have increased the availability of biologic therapies for inflammatory bowel disease (IBD), making it difficult for patients with ulcerative colitis (UC) and Crohn’s disease (CD) to navigate treatment options. Here we developed a conjoint analysis to examine patient decision-making surrounding biologic medicines for IBD. We used the results to create an online patient decision aid that generates a unique “preferences report” for each patient to assist with shared decision-making with their provider.
We administered an adaptive choice-based conjoint survey to IBD patients that quantifies the relative importance of biologic attributes (e.g., efficacy, side effect profile, mode of administration, and mechanism of action) in decision making. The conjoint software determined individual patient preferences by calculating part-worth utilities for each attribute. We conducted regression analyses to determine if demographic and disease characteristics (e.g., type of IBD and severity) predicted how patients made decisions.
640 patients completed the survey (UC=304; CD=336). In regression analyses, demographics and IBD characteristics did not predict individual patient preferences; the main exception was IBD type. When compared to UC, CD patients were more likely to report side effect profile as most important (odds ratio (OR) 1.63, 95% confidence interval (CI) 1.16–2.30). Conversely, those with UC were more likely to value therapeutic efficacy (OR 1.41, 95% CI 1.01–2.00).
Biologic decision-making is highly personalized; demographic and disease characteristics poorly predict individual preferences, indicating that IBD patients are unique and difficult to statistically categorize. The online decision tool resulting from this study (www.ibdandme.org) may be used by patients to support shared decision-making and optimize personalized biologic selection with their provider.
Inflammatory bowel disease (IBD) is a chronic, disabling condition that leads to significant morbidity and decreased health-related quality of life (HRQOL) (1, 2). There are many treatment options for ulcerative colitis (UC) and Crohn’s disease (CD), but biologic therapies remain the most effective and are a mainstay of treatment for those with moderate-to-severe IBD (3, 4). They can alter the disease course of UC and CD and achieve sustained mucosal healing, leading to symptom remission, fewer hospitalizations, and improved HRQOL (3, 4, 5, 6). As of October 2017, seven biologics are approved for use in IBD—infliximab, adalimumab, certolizumab pegol, golimumab, natalizumab, vedolizumab, and ustekinumab—and additional therapies are in the pipeline (7, 8).
While the available biologics are all effective in treating IBD compared to placebo, there have been no major head-to-head trials. Because of the lack of comparative effectiveness data, UC and CD care pathways endorse several first-line therapies for patients at increased risk for disease complications (9, 10). Adding to the complexity is the substantial variation among biologics with respect to mechanism of action, mode of administration, and side effects, among other attributes. For example, the available therapies can be categorized as anti-tumor necrosis factor (TNF), anti-integrin, or anti-interleukin (IL)-12/anti-IL-23 agents (7, 8). Aside from the mechanism of action, they differ in both the route (intravenous vs. subcutaneous) and frequency of administration. They also have varying side effect profiles, as there are differential rates of fatigue, skin rash, lymphoma, and serious infections (11). As a result, it is often difficult for patients to navigate the array of treatment options with their physician and to choose a therapy that aligns with their unique treatment preferences. Moreover, the decision-making process will become more complex as new drugs are developed and approved.
The principles of shared decision-making encourage better alignment of medical care with patients’ unique preferences and values (12, 13). In the case of IBD, shared decision-making requires that physicians understand how their patients decide among available treatments. In theory, prescribing a biologic that precisely maps to a patient’s preferences may yield improved adherence, better quality of life, enhanced clinical outcomes, and lower healthcare costs compared to a less-personalized approach of assigning therapy (14, 15, 16, 17).
In this study, we used conjoint analysis, a technique that determines how respondents make complex decisions under conditions of uncertainty, to examine how patients with IBD navigate among available biologic therapies. By examining patients’ choice patterns without reference to generic or brand names, we quantified and rank ordered the relative importance of biologic attributes (i.e., mechanism of action, mode of administration, efficacy, and side effect profile) that drive decision making. Further, we sought to characterize the heterogeneity among patients regarding preferences for biologic therapies, and to evaluate whether any demographic or disease characteristics may predict preferences. We used the resulting conjoint analysis to create a freely-available online patient decision aid, developed through collaborative and educational grants, that generates a unique “preferences report” for each patient to assist with shared decision-making with their provider.
Overview of conjoint analysis
Conjoint analysis is a form of tradeoff analysis that elucidates how people make complex decisions by balancing competing factors (18). It poses a series of side by side comparisons of competing product profiles and asks respondents to select which profile is preferable (Figure 1). On the basis of the respondent’s answer to the first comparison, an algorithm selects a new side by side comparison and asks the respondent to select the preferred profile. The process continues until the respondent reveals internal consistency and the technique collects sufficient data to rank preferences.
A task force from the International Society for Pharmacoeconomic and Outcomes Research (ISPOR) developed guidelines for the use of conjoint analysis in analyzing clinical decisions (18). Recent studies used the technique in evaluating clinical decision-making in rheumatology (19), spinal surgery (20), diabetes management (21), use of transfusions in dialysis-related anemia (22), and IBD (23, 24, 25, 26, 27, 28), among many other applications. Conjoint analysis is now considered to be a robust approach to analyze decision making in healthcare (18).
Adaptive choice-based conjoint analysis for biologic decision-making
To quantify and rank preferences regarding use of biologics in IBD, we used the adaptive choice-based conjoint (ACBC) platform developed by Sawtooth Software (Sawtooth, North Orem, Utah). The ACBC approach assumes that decision making depends upon attributes, each of which has levels. For example, selecting among IBD biologics may depend upon many attributes, including efficacy, route and frequency of administration, lymphoma risk, serious infection risk, tolerability of side effects, risk of developing a rash, fatigue, and mechanism of action. Each attribute can be measured across several levels. See Table 1 for the nine biologic attributes and associated levels tested in our survey and Supplementary File A online for the survey text.
Once the attributes and levels are defined, the ACBC software displays sets of side by side therapeutic profiles, each with varying levels for each attribute (Figure 1). For each therapeutic profile in these “screener tasks,” respondents decide whether the therapy is acceptable. The comparisons become increasingly complex as the respondent progresses. In order to tailor the survey to each respondent’s unique preferences, the software also asks about attribute level extremes, including “unacceptable” vs. “must-have” rules that should be incorporated in subsequent choice tasks. Next, the ACBC software guides respondents through a “choice tournament” that presents sets of side by side therapeutic profiles based on earlier selections (Figure 1); the respondents choose which is preferred and continue until responses achieve internal consistency.
After respondents complete the ACBC survey, the conjoint software uses hierarchical Bayes regression to estimate individual-level utility coefficients by iteratively estimating and borrowing information from similar respondents (24, 29). The utility coefficients are called part-worth utilities, and they are generated for each attribute level. Attribute levels that have greater importance in the decision-making process are associated with higher part-worths. The part-worth utilities for the levels within each attribute sum to zero.
In addition to calculating part-worth utilities, the ACBC software also generates importance scores. The importance scores are derived by calculating the delta between the part-worths for the most important and least important level of each attribute (24). The larger the delta in part-worth utilities, the larger the importance of the attribute in the decision-making process. Refer to Supplementary File B for more information about ACBC, part-worth utilities, and importance scores.
We created the conjoint survey using input from the literature (23, 24, 28, 30, 31), as well as findings from our IBD social media netnography research (32), to ensure content coverage of factors related to patient biologic decision-making in IBD. We tested nine attributes organized in four categories: (i) biologic mechanism of action; (ii) mode of administration (i.e., route and frequency); (iii) efficacy—long-term remission, short-term improvement; and (iv) side effect profile—tolerability of side effects, fatigue, rash, risk of serious infection, risk of lymphoma (Table 1).
In addition to conjoint vignettes, the survey included standalone questions regarding patient demographics, including age, gender, race/ethnicity, education, marital status, employment status, and income. We also asked questions regarding participants’ IBD, including the type of IBD (UC or CD), duration of IBD, prior IBD-related surgery, IBD-related symptoms experienced in the past 30 days, IBD severity as determined by the Short Inflammatory Bowel Disease Questionnaire (SIBDQ) (33), and current and prior IBD therapy use (steroids, aminosalicylates, immunomodulators, antibiotics, and biologics). Responses to these questions were used to identify potential demographic or clinical predictors of decision making. We hypothesized that biologic preferences might vary predictably, for example, by IBD severity or patient age. This study was approved by the Cedars-Sinai Institutional Review Board (Pro39038).
Participants with IBD were recruited through three research panels between February 2016 and September 2016 to complete the online survey. One panel consisted of IBD patients identified through the National GI Survey—a population-based audit conducted to measure gastrointestinal (GI) symptoms in community-dwelling Americans (34, 35, 36). All respondents were asked about GI symptoms they recently experienced along with relevant demographics and comorbidities including UC and CD. Among the 71,812 individuals who completed the National GI Survey, 494 (0.7%) and 506 (0.7%) reported having been diagnosed by a physician with UC and CD, respectively.
Because it is possible that the National GI Survey cohort may include false positive IBD diagnoses, we bolstered recruitment using two databases comprised of individuals whose IBD was medically confirmed by a gastroenterologist. We subsequently performed sensitivity analyses, described below, to evaluate for evidence of response differences between groups. For medically confirmed IBD patients, we recruited from the GI Patient Reported Outcome Measurement Information System (PROMIS) research database (37) and the Cedars-Sinai Mucosal Immunology Repository for Inflammatory and Digestive Diseases (MIRIAD) database. The PROMIS database includes 2042 general GI patients evaluated in clinics at Cedars-Sinai Medical Center, the West Los Angeles Veterans Affairs Medical Center, the University of Michigan, and the University of California at Los Angeles. Of this group, 56 (2.7%) and 97 (4.8%) have medically confirmed UC and CD, respectively. The MIRIAD database includes email addresses for 250 UC and 529 CD patients.
We recruited individuals, 18 years of age or older, with evidence of recently active IBD symptoms in the past 30 days, including abdominal pain, diarrhea, constipation, bowel incontinence or leakage, nausea/vomiting, joint pain, or blood in the stool. Of note, we did not have access to laboratory, radiographic, or endoscopic data, and could not definitively determine whether individuals had active disease.
Sample size and statistical analyses
On the basis of conjoint analysis sample size precedents and the recommendations of the software provider (38), we aimed to recruit 300 UC and 300 CD patients to complete the conjoint analysis survey. For the entire study cohort, descriptive analyses were performed to calculate means and proportions for demographic and IBD characteristics. In addition, we calculated mean importance scores for each biologic attribute, and listed them in rank order from highest to lowest relative importance. We then evaluated group-level (e.g., UC vs. CD) rankings, followed by patient-level “preferences report” ratings to assess the uniqueness of individual’s decision profiles.
We performed multivariable logistic regression models to adjust for potentially confounding factors and to calculate odds ratios (ORs) and 95% confidence intervals (CIs). The outcomes in the regression models were whether individuals reported the following attribute categories as the most important factor in their decision making: (i) mechanism of action; (ii) mode of administration; (iii) efficacy (long-term remission or short-term improvement); and (iv) side effect profile (tolerability of side effects, fatigue, rash, risk of serious infection, or risk of lymphoma). All patient-level demographic (age, gender, race/ethnicity, education, marital status, employment status, household income) and clinical (type of IBD, duration of IBD, prior surgery for IBD, IBD severity as determined by the SIBDQ, current and prior IBD therapy use) variables were included in the regression models. Sensitivity analyses were also conducted among the following subgroups of patients: (i) medically-confirmed IBD patients; (ii) biologic-naive; and (iii) biologic-experienced. Statistical analyses were performed using Stata 13.1 (StataCorp LP, College Station, TX). A two-tailed P-value <0.05 was considered significant.
Development of online shared decision-making tool
We translated the conjoint analysis results into a clinically useful, freely available online tool for shared decision-making. The resulting web-based self-assessment site, called “IBD&me,” allows patients to explore decision making around biologic therapies for IBD. Funding for this website was provided through collaborative research and educational grants from Takeda Pharmaceuticals to the Cedars-Sinai Office of Continuing Medical Education (CME), which oversaw development of the content using a fair and balanced, peer-review process.
The purpose of IBD&me (www.ibdandme.org) is to enable patients to explore biologic risks and benefits and to objectively assess latent treatment preferences. We worked with user interface and experience experts (Objectiva; www.objectivasoftware.com) to create an unbranded mobile-compatible website including CME for clinicians (cme.ibdandme.org), patient education, and a patient assessment tool (called the “IBD&me Decision Tree,” powered by the Sawtooth Software conjoint analysis used in this study). The website generates a unique “preferences report” for each patient (example in Figure 2) depicting their importance scores and rankings. Of note, we are conducting formal usability testing of IBD&me through in-depth cognitive interviews with IBD patients, which will inform iterative updates to the site.
Supplementary Figure 1 displays a flow diagram of enrolled patients. Of the 800 UC and 1,132 CD patients who were invited to participate, 304 (38.0%) and 336 (29.7%) with UC and CD, respectively, completed the conjoint analysis. Table 2 presents the demographic and clinical characteristics of the 640 IBD patients included in the final analyses.
Overall rank ordering of biologic attribute importance
When grouping the nine attributes into four overarching categories, the ACBC algorithm revealed that 264 (41.3%) respondents valued treatment efficacy as the most important attribute category in their decision making, followed by 245 (38.3%) for side effect profile. In contrast, 128 (20.0%) and 3 (0.5%) individuals valued mode of administration and mechanism of action, respectively, as the predominant factor.
Figure 3 depicts the average utility-derived importance of the nine individual biologic attributes, stratified by IBD type. On average, the ACBC algorithm revealed that UC patients valued long-term remission rates, mode of administration, lymphoma risk, and short-term improvement rates as the top four most important factors. Conversely, for CD patients, short-term improvement rates, lymphoma risk, mode of administration, and long-term remission rates were the top four factors. For both UC and CD patients, rash, fatigue, and biologic mechanism of action were less important (see Figure 3 for all importance scores).
Intravenous vs. subcutaneous mode of administration
Overall, part-worth assessment revealed that 295 (46.1%) individuals preferred intravenous administration of the biologic in a clinic while 345 (53.9%) favored administering it themselves subcutaneously at home. In logistic regression analyses, we found that non-Hispanic blacks (n=17, 68.0%; OR 3.22, 95% CI 1.32–7.86) and Latinos (n=61, 56.5%; OR 1.67, 95% CI 1.07–2.61) had significantly higher odds of preferring the intravenous route when compared to non-Hispanic whites (n=198, 42.2%; reference). The remaining demographic and IBD variables in the model were not predictive of route preference.
Uniqueness of individual “preferences report”
For each respondent, the conjoint software rank ordered the relative importance of the nine biologic attributes, effectively presenting the results as an individual “preferences report.” The report lists both the absolute rank of biologic attributes in order of preference, and the relative weight of each attribute. Both types of information are important, as understanding the absolute and relative importance is often more helpful than knowing unweighted rank order alone.
Comparison of preferences reports among participants revealed that individual patients were unique—there were few patients with similar decision-making profiles. Among all 640 respondents, no two shared the same report in terms of both absolute and relative attribute importance. When focusing only on absolute preference ranking, there was some overlap but still a high degree of uniqueness among patients. Figure 4 displays the proportion of unique decision-making profiles according to attribute ranking. For example, when examining the rank ordering of all nine attributes, 97.7% of the 640 patients had unique reports. When limiting our analysis only to the top four attributes, there was still uniqueness in 46.9% of patients. Not until limiting our analysis to the top three attributes did we observe more overlap in patient preferences.
Predictors of biologic decision-making
Table 3 presents the results of multivariable logistic regression analysis testing for independent predictors of patient preference. Overall, despite a relatively large sample size and multiple comparisons, we found few demographic or clinical predictors of patient preference, again indicating that the sampled IBD patients were unique and difficult to statistically categorize. However, there were scattered independent predictors, as follows.
Mode of administration
The only variable predicting that a patient would endorse mode of administration as the most important attribute category was race/ethnicity; non-Hispanic blacks (OR 2.55, 95% CI 1.05–6.20) were more likely to value this attribute vs. non-Hispanic whites. Among the 128 individuals who prioritized route/frequency of administration, 58 (45.3%) and 70 (54.7%) preferred intravenous and subcutaneous delivery, respectively.
Regarding biologic efficacy, individuals with UC were more likely than those with CD to value efficacy as the most important attribute category (OR 1.41, 95% CI 1.01–2.00), holding all other variables constant. Patients with prior IBD surgery were also more likely to report efficacy as predominant (OR 1.62, 95% CI 1.11–2.37). Conversely, increasing age was associated with decreased odds for valuing efficacy as the primary determinant (18–30 years—reference; 31–45 years—OR 0.63, 95% CI 0.41–0.96; ≥46 years—OR 0.58, 95% CI 0.36–0.96). The remaining demographics and IBD characteristics largely were not independently predictive.
Side effect profile
Patients with CD were more likely than those with UC to report avoidance of a biologic side effect as the primary determinant when choosing among biologics, all else equal (OR 1.63, 95% CI 1.16–2.30). The other demographics and IBD characteristics were not predictive.
Medically confirmed IBD patients
When focusing only on individuals confirmed to have IBD by a gastroenterologist, we found that the overall attribute rank order was similar compared to the primary analyses. Namely, most respondents (n=51, 52.6%) still reported biologic efficacy as the most important determinant of decision making. Thirty-one patients (32.0%) noted that avoidance of a side effect was the primary determinant when choosing among options, while 15 (15.5%) reported that mode of administration was the most important factor. No individuals prioritized mechanism of action in their decision making.
The regression analyses were largely unchanged in sensitivity analyses. Those with CD remained more likely (OR 8.30, 95% CI 1.33–51.63) to report side effect profile as the most important factor vs. UC patients. Those with UC also still had higher odds (OR 6.15, 95% CI 1.36–27.82) for reporting efficacy as the most important factor compared to individuals with CD.
Biologic-naive vs. -experienced IBD patients
Among patients who have never been on a biologic (n=309), efficacy remained the priority for most individuals (n=126, 40.8%), followed by side effect profile (n=122, 39.5%) and then mode of administration (n=61, 19.7%). No individuals reported mechanism of action as the most important factor. Biologic-experienced patients (n=331) had a similar distribution: (i) efficacy—n=138, 41.7%; (ii) side effect profile—n=123, 37.2%; (iii) mode of administration—n=67, 20.2%; and (iv) mechanism of action—n=3, 0.9%.
In regression analyses among biologic-naive individuals, we found that those with CD were still more likely (OR 1.95, 95% CI 1.21–3.16) to prioritize side effect profile, while those with UC remained more likely to value efficacy (OR 1.93, 95% CI 1.17–3.16) when selecting among the different options. Non-Hispanic blacks also still had higher odds for prioritizing mode of administration (OR 4.93, 95% CI 1.49–16.35). In contrast to the primary analysis, prior surgery (OR 1.30, 95% CI 0.69–2.44) and increasing age (18–30 years—reference; 31–45 years—OR 0.66, 95% CI 0.35–1.25; ≥46 years—OR 0.57, 95% CI 0.28–1.16) were no longer significant predictors of reporting efficacy as the dominant factor in decision making. On the other hand, those with more severe IBD (first quartile SIBDQ scores) had 2.26 (95% CI 1.01–5.05) times the odds of prioritizing medication efficacy vs. those with less severe disease (fourth quartile SIBDQ scores).
When focusing on biologic-experienced patients, CD and UC were no longer significantly associated with reporting side effect profile and efficacy, respectively, as the most important factors when choosing a therapy. In fact, among this subgroup, demographics and IBD characteristics largely were not predictive of decision making, save for a few exceptions. Namely, those who had prior surgery were more likely to value efficacy (OR 1.69, 95% CI 1.03–2.79). We also found that more educated individuals had higher odds for reporting efficacy as the most important factor (high school or less—reference; some college—OR 2.86, 95% CI 1.22–6.74; college degree—OR 2.16, 95% CI 0.93–5.06; graduate degree—OR 2.70, 95% CI 1.03–7.03).
Using conjoint analysis, we found that the determinants of biologic therapy decision-making vary between patients with UC and CD. Whereas UC patients value biologic efficacy as the most important factor when deciding among biologics, patients with CD prioritize side effect profile. This suggests that UC and CD patients approach biologic decision-making in a systematically different manner.
Moreover, among the sampled IBD patients, we found widely divergent individual patient preferences when selecting among IBD biologics. In attempting to identify predictors of individual choices, we found that demographic and IBD characteristics are largely unhelpful, save for a few exceptions. For example, older patients are less likely to value efficacy as the single most important attribute, while those with prior IBD surgery are more likely to value efficacy. We also found that non-Hispanic blacks are more likely to prioritize mode of administration vs. non-Hispanic whites—a result of unclear significance that may be spurious, particularly given that black patients were underrepresented in this study. Other than these few results, the remaining patient demographic characteristics are not predictive of decision making. Similarly, IBD duration, severity, and IBD medication history are largely unrelated to patient priorities when navigating the array of biologic treatment options. When considering each respondent’s individual “preferences report,” we found that 98% had a completely unique decision-making profile. These results emphasize that biologic decision-making is highly individualized; providers cannot rely on demographic or clinical variables to neatly categorize patient preferences or attempt to predict which biologic will optimal map with a patient’s personal values.
Because of the highly individualized nature of decision making in IBD, along with healthcare’s increased emphasis on shared decision-making, it is critical for clinicians to identify what matters most to patients when choosing among therapeutic options; this enables patients to select therapies that align with their values. Yet, it can be challenging to accurately establish a patient’s unique preference profile in the context of a brief clinical visit. In the face of burgeoning administrative and clinical tasks, gastroenterologists often lack enough time and resources to engage in detailed discussions around biologic risks, benefits, and tradeoffs. Thus, there is a need for simple and efficient decision tools that elicit individual preferences and support the patient-provider interaction.
To address this gap, we used the conjoint analysis developed and tested in this study to support an online decision aid called IBD&me (www.ibdandme.org). The website uses conjoint analysis to quantify and rank the biologic attributes that drive an individual patient’s decision-making preferences. After patients complete the assessment, the decision tool generates a personalized “preferences report” that displays both the absolute rank order and relative weights of attributes that matter most to an individual (see Figure 2 for an example). The patient can share the report with his or her providers, who can review the information to quickly understand the patient’s preferences and values. The information may also allow providers to identify what discussion topics to focus on during the visit. For instance, if a patient strongly values avoiding side effects at all costs, then the provider should take special care to place biologic risks into context, particularly since research indicates that IBD patients tend to misperceive the risks of therapy (31, 39). If the report shows, for example, that lymphoma risk is the single most important factor for a patient, then the provider should consider using a Paling Palette or similar visualization tool to ensure lymphoma risk is clearly explained (40).
By helping patients and providers have more informed and meaningful discussions in clinic, the tool may lead to improved shared decision-making, enhanced satisfaction with the visit, and higher medication adherence. Nonadherence to maintenance medications, including biologics, is estimated to occur in 30–45% of IBD patients and leads to increased relapse rates as well as higher overall healthcare costs (41). However, the shared decision-making literature indicates that enabling more personalized treatment choices may increase adherence with treatment plans (41, 42). In our future research, we aim to test and validate whether the IBD&me decision tool improves patient satisfaction, shared decision-making, biologic adherence, and ultimately IBD outcomes when compared to usual care.
Our study has several strengths. First, it is one of a small number that have used conjoint analysis to examine IBD patient decision-making (23, 24, 25, 26, 27, 28). For example, Lichtenstein et al. (24) previously employed conjoint analysis to examine patient preferences regarding biologics, but their study did not include patients with UC and did not include side effects among the attributes, a preeminent concern for many patients. In contrast, we included a wide range of biologic attributes that patients consider when deciding among treatment alternatives, and tested decision making in a larger sample including both UC and CD patients. Second, we performed regression analyses to identify potential predictors of decision-making. We found very few independent predictors of individual patient preference, suggesting that providers cannot easily surmise what tradeoffs patients are willing to take solely based on demographic and clinical data available in the medical chart; understanding unique patient preferences requires a more sophisticated approach. Moreover, while prior research found that IBD patients, on average, are willing to tolerate elevated risks of serious adverse events in exchange for therapeutic benefit (23), our results suggest that individuals differ widely on just how much risk—if any—they find acceptable.
Our study also has important limitations. First, in a conjoint exercise, participants are required to make tradeoffs between hypothetical treatments (albeit based on attributes of available biologics), and it is possible that subjects may behave differently than if they were making a real-world decision. Directly observing patient-provider interactions is considered the gold standard for assessing process of care. However, direct observation is also limited because of the Hawthorne effect in which patients and providers may alter their practice when they are knowingly observed. Standardized patients (43) and medical record data abstraction (44) are alternatives. Notably, survey-based clinical simulations have been previously validated as an accurate surrogate for both chart abstraction and standardized patients (45), and are thus widely recognized to be a valid, reliable, practical, and cost-effective technique to assess process of care and clinical decision-making among patients and providers. Second, our survey featured a limited number of attributes that cannot capture the full breadth and depth of patient decision-making surrounding biologics, such as ability to adjust therapies to optimize outcomes using therapeutic drug monitoring, length of time since drug approval, avoidance of surgery, etc. However, this was by design, as ACBC surveys can become unwieldy with too many attributes. Sawtooth Software recommends minimizing the number of included attributes to a core set (46); we chose to include nine attributes to keep the survey length at about 15–20 min. Moreover, we leveraged our previous social media netnography research to determine attributes that weighed heavily in IBD patients’ decision making around biologics (32).
A third limitation is that it may be cognitively challenging for some respondents to synthesize information across many different attributes when comparing treatments. However, this is comparable to how patients currently make decisions in a clinical setting. In fact, it can be argued that conjoint is superior in eliciting patient risk-benefit preferences, as patients can complete the survey at their own pace and spend more time carefully considering tradeoffs. Similarly, due to software limitations that restricted how graphics are displayed, we did not include visualizations (e.g., Paling Palettes) to more clearly represent risks in the conjoint exercise. Participants with lower numeracy skills may have struggled with parts of the survey, being more prone to denominator neglect or having difficulties switching between percentages and fractions. Again, this mirrors what occurs in clinical practice, particularly during brief clinical visits when visualization tools are not utilized. Our IBD&me decision aid attempts to address potential numeracy issues by including graphics and videos that employ Paling Palette-like visualizations in the “Learn More” section to allow patients to better understand the lymphoma and serious infection risks related to biologics.
A fourth limitation is that our sample included a group of self-reported IBD patients in addition to those with a confirmed diagnosis. It is possible that some of those with self-reported IBD did not truly have IBD. However, a prior study of an Internet-based IBD patient cohort from CCFA Partners found that self-reported IBD diagnosis is highly accurate, as 97% were validated to have physician-confirmed IBD (47). Moreover, in our study, we performed a sensitivity analysis including only those with medically confirmed IBD, and the results were largely unchanged when compared to the primary analyses.
Finally, our survey response rate of 33.1% was relatively low, leading to potential selection bias that may limit the generalizability of our results. However, our response rate was comparable to other published studies using conjoint analysis that recruited IBD patients through Internet panels (23, 24). It is also unlikely that survey non-responders approach selection of a biologic in a systematically different and predictable manner when compared to responders, especially in light of the highly-individualized nature of decision making seen in our study.
In summary, we found that the determinants of biologic therapy decision-making vary between patients with UC and CD. We also found that the decision-making process is highly personalized and that demographic and IBD characteristics poorly predict individual patient preferences. It is increasingly difficult for patients to select among the growing number of IBD biologics in partnership with their healthcare providers. The IBD&me online decision tool emerging from this research may help improve shared decision-making and optimize biologic selections in a more personalized and structured manner than often allowed by time and resource constraints.
We would like to thank and acknowledge Loren C. Karp for her project management and support.
SUPPLEMENTARY MATERIAL is linked to the online version of the paper at http://www.nature.com/ajg