Structured Abstract

Background

Inflammatory bowel diseases (IBDs), including Crohn's disease (CD) and ulcerative colitis (UC), affect more than 1.5 million Americans and 3 million Europeans, with peak incidence in the second and third decades of life.^1-6 Importantly, CD, in general, and UC, when severe, are associated with increased mortality.^7,8 Both diseases are associated with intermittent symptoms such as diarrhea and abdominal pain, reduced productivity in work and school, intermittent need for hospitalizations, and, as a consequence, reduced quality of life.^9-11

Numerous medications are efficacious in the treatment of IBD, including traditional corticosteroids (CS), budesonide, thiopurines, methotrexate, anti-tumor necrosis factor α (anti-TNF) therapies, and other biologics (natalizumab, vedolizumab, and ustekinumab).^12-15 Traditionally, treatment strategies for both CD and UC have used the least effective, but perceived safest, therapies initially, reserving therapies that are more effective but perceived as potentially more toxic (ie, anti-TNF, thiopurines, methotrexate, and natalizumab) for patients with more refractory disease.^12,16 Typically, this has followed a step-up algorithm, whereby mesalamine and/or CS are used initially, and if remission is unable to be maintained without CS, then the addition of a CS-sparing agent such as a thiopurine or anti-TNF drug would be added, sometimes in combination. Recent data have prompted consideration of earlier use of the more effective therapies, with the goal of preventing long-term complications such as the need for an ostomy or repeated surgeries.^12,17 In general, the efficacy of therapies for UC parallel that of CD. The most notable difference is that mesalamine and CS are more efficacious in UC than in CD,^18-21 which means that fewer patients with UC require escalation of therapy to thiopurines and/or anti-TNF.

Whether intermittent immunosuppression with CS or chronic immunosuppression with anti-TNF is the preferred strategy has not been definitively answered. Randomized trial data that directly compare a strategy of intermittent CS therapy vs induction and maintenance anti-TNF therapy are limited.¹⁷ However, the evidence base for efficacy is greater for anti-TNF drugs^22-25 and these drugs sometimes allow for CS discontinuation.^26-28 Nonetheless, CS, alone or in combination with other medications, remain commonly used to treat IBD, in part due to fear of adverse effects from anti-TNF drugs.^29-31

Serious adverse events differ between intermittent CS use and anti-TNFs. Examples of uncommon but potentially fatal adverse reactions that have been associated with CS and/or anti-TNF include serious infections (both drug classes),^32-37 congestive heart failure (both drug classes),^38,39 cancer (with anti-TNF),^40,41 osteoporosis and fractures (with CS),^42,43 and pulmonary embolus (with CS).⁴⁴ However, poorly controlled disease may also increase the risk of death and serious complications, further complicating the choice between therapies.^8,35

An increased risk of death has been observed among patients taking CS for IBD.^8,45 The relationship between anti-TNF therapy and mortality is less clear. A 32% lower mortality rate (not statistically significant) was observed in a group randomized trial comparing protocol-driven early combined immunosuppression with an anti-TNF drug plus an immunomodulator drug vs usual care for patients with symptomatic CD.⁴⁶ In contrast, in a retrospective cohort study, Fidder et al observed a 1.9-fold increased, but not statistically significant, risk of death with anti-TNF therapy.^47,48 The TREatment of ATopic eczema (TREAT) registry, composed largely of prevalent infliximab users, did not observe an increased or decreased risk of death with infliximab. However, the TREAT study may be subject to bias from depletion of susceptible subjects because it did not focus on new users of anti-TNF therapy (ie, prevalent cohorts study patients who have survived initial therapy and are more likely to have benefited from therapy).⁴⁵ To our knowledge, no study has specifically compared mortality rates among anti-TNF treated patients to that of CS-treated patients or examined key subpopulations such as the elderly or those with more comorbid illnesses who might preferentially benefit from one therapy vs another.

Historically, patients had relatively little input into treatment strategies. Rather, regulators approved drugs for marketing and physicians chose to whom to prescribe the drug. However, the practice of medicine in the 21st century involves shared decision-making. Even the most well-intentioned physician cannot treat a patient with the recommended therapy if the patient is unwilling to take the therapy due to personal preferences. The degree to which patients' preferences contribute to the previously described variability in the care of patients with IBD has not been well studied.

To address how a patient makes treatment preferences for inflammatory bowel disease and whether anti-TNF therapy or prolonged CS therapy is associated with greater overall mortality and quality of life, we designed a 3-part study. In Phase 1, we used latent class analysis of choice-experiment data obtained from a web-enabled survey to quantify IBD patients' preferences for duration of remission, steroid use, and risk of key adverse outcomes including lymphoma, serious infection, and surgery. Differences in observed choice patterns indicated the implicit rates at which respondents were willing to accept trade-offs among benefits and risks. We used these estimates to compute remission time equivalents (RTEs), a novel utility metric for inflammatory bowel disease, for relevant health states. In Phase 2, we conducted a comparative effectiveness study of CS vs anti-TNF therapy for IBD with death as the primary outcome and select serious adverse events as secondary outcomes. Finally, in Phase 3, we combined the RTEs from Phase 1 with the data from the retrospective cohort study in Phase 2 to compare quality of life between the 2 therapies and build a mathematical Markov model to mimic the cohort study, allowing for assessment of the robustness of the results to assumptions about the probability of different outcomes and the RTE estimates. The data derived from these 3 phases of investigation demonstrate that the use of anti-TNF therapy as a steroid-sparing strategy for patients with CD results in improved survival, reduced rates of several life-threatening complications, and improved quality of life. For patients with UC, the trends in reduction in mortality with anti-TNF therapy were in the same direction, but the magnitude of benefit was smaller.

Methods

Phase 1—Discrete Choice Experiment

Survey Sample

We sent invitation emails to members of the Crohn's and Colitis Foundation of America (CCFA) Partners cohort with self-reported diagnoses of CD or UC.¹³ CCFA Partners is a cohort of more than 15 000 adult patients with IBD who have agreed to complete online surveys related to their disease.⁴⁹ The cohort was established in 2012 and prior studies have demonstrated a 97% accuracy of the self-reported IBD diagnosis.⁵⁰ Members of the CCFA Partners cohort complete semiannual online surveys related to their disease. Following completion of the semiannual questionnaire, participants were invited to complete an additional survey about their preferences for therapies for their IBD. Those who agreed were emailed a separate invitation that described the questionnaire and included a link to access the online survey.

Survey Development

We developed a choice-experiment survey using best-practice methods to elicit patients' willingness to accept trade-offs among various medication and surgical therapies.¹⁴ Baseline demographics and recent disease history from the CCFA Partners' database were augmented by additional questions about specific history related to the risks and therapies assessed in the study.

Treatment attributes shown in choice-experiment scenarios were determined from literature review, IBD expert consultation, focused interviews with IBD patients, and piloting in 9 Crohn's disease patients. Patients were asked to assume that their current treatment to control their IBD was not working and they needed to consider alternative therapies. Patients with CD were offered choices between medical therapy treatment A and treatment B. Participants were informed that these hypothetical treatments did not necessarily represent currently available therapies. Attributes of each treatment included number of months per year of specified disease severity ranging from 12 months of remission to 12 months of mild, moderate, or severe disease activity, and number of months of steroid usage each year (Figure 1.1). Symptom descriptions were adapted from the Crohn's Disease Activity Index.

Figure 1.1

Attributes and Levels.

Three potential serious adverse events also were considered. The increased risk of lymphoma, serious infection, and, for CD, the need for intestinal surgery associated with each treatment was described in nontechnical language. Hypothetical risk levels for a 1-year period ranged from 0% to 8% for lymphoma and surgery and 0% to 30% for serious infections. Pretest interviews and pilot data indicated upper limits required to quantify the maximum levels of risk patients would accept for defined levels of benefit. To limit cognitive burden and numeracy concerns, all treatment benefits were described as certain and all treatment risks were described as known probabilities. Consistent with best practices, specific risk levels were shown (rather than ranges, to avoid measurement error).¹⁴ To further assist respondents in understanding quantitative risks, serious adverse event probabilities were presented both graphically—in a risk icon array of shaded humanoid figures indicating the number of patients out of 100 who would have the serious adverse event—and numerically—as counts out of 100 and percentages (Figure 1.2).¹⁵

Figure 1.2

Example of Survey Scenario.

We used a variation of a commonly used algorithm in SAS to construct an experimental design that resulted in 36 pairs of treatment options.^16-20 To reduce respondent burden, we divided trade-off scenarios into 4 survey versions of 9 questions each. Each participant was randomly assigned to receive 1 of the versions. We emailed surveys using the Dillman method to maximize response rates.²¹

Survey Validation

The choice-experiment surveys included tests for numeracy and an internal test for subject-level validity through logic testing. To assess understanding of the survey's numerical concepts, subjects were shown a series of numerical examples of risk—presented as percentages, fractions, and a risk-grid graphic—and subsequently tested on their understanding of these numeric concepts. We assessed logic testing to evaluate if respondents understood the question-choice format sufficiently to indicate a preference for a better therapy through a trade-off scenario in which 1 medication treatment dominated the alternative for every attribute. We tested the model to evaluate the statistical influence of respondents who failed 1 or both tests.

Statistical Analysis

We evaluated responses to the dominated-pair choice question for internal validity and the numeracy quiz, where correct answers indicated attentiveness and understanding of the survey content.

We used Latent GOLD 5.0 Choice to estimate latent class relative importance weights from the choice data.²² The choice model estimates separate parameters and class membership probabilities for a specified number of classes. We estimated models for 1, 2, 3, 4, and 5 classes to evaluate the optimal number of classes using procedures discussed in Lanza and Rhoades.⁵¹ We also evaluated parameter estimates for theoretical interpretability and clinical relevance. We tested effects-coded log-odds parameter estimates for significant differences across latent classes using Wald statistics, and we used bivariate statistics to test for differences in covariate estimates across classes. We selected the 3-class model.

We used the bayesian information criterion for model fit, as well as interpretability and clinical plausibility, to assess results across models. We specified fully categorical models to avoid imposing functional-form assumptions for continuous variables. We estimated all severity-duration interactions to avoid the implausible, but common, assumption that health-state utility and durations are linear and proportional. We used Wald tests to determine whether coefficient differences were significant among classes.

The absolute scale of the preference-utility parameter estimates have no intuitive meaning: Only comparisons of differences are meaningful. Thus, we calculated the relative importance of each attribute over the range evaluated by dividing the absolute differences between the largest and smallest parameters for each attribute by the sum of the absolute differences between the largest and smallest parameters for all the attributes. The relative importance score indicates the overall influence each attribute had on choice evaluations.

We also obtained a standardized metric for utility differences by scaling differences by the marginal utility of 1 month of remission to derive the RTE of a given duration of symptom severity. We specified time profiles as number of months with specified symptom severity and remaining number of months of remission over a 12-month period. Thus, the utility of an additional month of remission is the utility gained from 1 less month of specified severity. We used the mean value over all severity-duration levels to scale severity durations. We define RTE for given duration of symptom severity as

$$RT{E}_{St}=\frac{U_{St}-U_{So}}{\Delta U_R}=\frac{{\beta }_{St}-{\beta }_{so}}{{\displaystyle \sum i}{\displaystyle \sum i}j\frac{{\beta }_{ij}-{\beta }_{io}}{t_{ij}}\text{N}ij}$$

where RTE_St is the healthy time equivalent of symptom severity S for duration t in months; U_So is the utility of no months of severity S; and U_St is 4, 8, or 12 months of symptom severity S estimates with categorical parameters β_St and β_So. The denominator ΔU_R is the mean marginal utility of 1 month of remission, which is calculated as the negative sum of all estimated severity utility differences per month divided by the total number N_ij of utility differences evaluated, where i indexes attributes and j indexes levels.

Deviation From the Research Proposal Submitted to PCORI

In the application to PCORI, the investigators proposed that subgroup analyses would be conducted among patients with and without a history of disability due to IBD, with and without a history of disability for other reasons, and with and without prior use of anti-TNF therapy. Unfortunately, in the process of developing the survey instrument for Phase 1, the investigators forgot to add the questions asking whether the participant receives disability benefits for IBD or for other reasons. As such, it was not possible to conduct these analyses as proposed in the research application. In addition, we modified the analysis based on use of anti-TNF therapy to be current use of any immunosuppressant drug, since current evidence suggests that the risks of infection and cancer are not unique to anti-TNF medications.

Phase 3—Preference-Weighted Cohort Study

In Phase 3, we combined the results of the Phase 1 discrete choice experiment with the cohort results for patients with CD from Phase 2 using 2 complementary approaches. In the first approach (referred to as the cohort study), for each patient with CD included in Phase 2, we computed an RTE-weighted value for each month of therapy using an initial treatment-carried-forward model that allows patients to switch therapies if the first is not effective. RTE-weighted values can be thought of as a quality-of-life value for each month of follow-up where the common reference is a month with Crohn's disease in remission. We mated the 2 treatment groups on propensity scores derived from the same covariates used in Phase 2 and compared the sum of the monthly RTE values between the groups using a paired t test. This determined which treatment group, on average, experienced greater quality of life over the follow-up period. In the second approach (referred to as the Markov simulation), we applied the transition probabilities derived from the cohort study and the RTEs from the preference experiment to a Markov simulation model and conducted sensitivity analyses to assess whether the results from the cohort study were sensitive to estimates of the transition probabilities or RTEs.

For both the cohort study and the simulation model, we applied RTEs to each month of follow-up according to the assigned health states. For example, if a patient had moderately active disease for the first 2 months of follow-up, was in remission for the next 8 months, and had a relapse with severe disease for months 11-12, then at the end of 12 months we assigned the patient the sum of 8 months of remission, 2 months of moderate disease, and 2 months of severe disease using the RTEs for each of these states computed in Phase 1.

To implement this analysis, it was necessary to assign a health state for each patient for each month by extrapolating from the Medicare and Medicaid claims data. This required certain assumptions that are summarized in Table 3.1. For time with metastatic cancer and nonmetastatic cancer treated with chemotherapy or radiation therapy, we extrapolated RTEs from published studies of traditional health-utility estimates.^68-75 We generally computed health states in monthly intervals, but in selected circumstances health states were assumed to extend for longer periods. For example, we assumed that 2 months were necessary for induction therapy with anti-TNF and CS. This is consistent with the induction dosing of the anti-TNF medications and the common practice of starting prednisone at 40 mg to 60 mg per day and tapering over several months. Whenever more than 1 health state could apply, we applied the one with the lowest RTE (most negative). In addition, several clinical events superseded mild, moderate, and severe states when they occurred, based on expert opinion. Postoperative states took precedence over any disease activity state but not necessarily over complications, such as metastatic cancer. Ostomy-related states took precedence over mild disease and remission, but not other states.

Table 3.1

Clinical Events and Assigned Remission Time Equivalents.

Analysis of the Patient Preference–Weighted Cohort Study

To compare mean quality of life estimates during follow-up, we matched individuals meeting criteria for prolonged CS exposure 1:1 to individuals within the anti-TNF arm via nearest neighbor caliper-based propensity score (PS) matching with 0.05 caliper. Variables used in the PS were the same as those used in Phase 2. Before matching, we excluded outliers with extreme PS values by trimming the tails of the PS distributions. We used SMDs to assess balance in the covariates among the anti-TNF and prolonged CD-treated patients in the matched cohorts. We used a SMD above 0.1 to define significant imbalance. We conducted all analyses as initial treatment carried forward after start of follow-up time even if the patient switched from CS to anti-TNF or vice versa. This model design tests the hypothesis that trying anti-TNF therapy, even if it is unsuccessful and discontinued, is more or less effective than a strategy of prolonged CS use. The primary analysis used a 12-month follow-up period. For patients who were lost to follow-up, such as for discontinuation of Medicare Part D benefits, before the end of follow-up, we computed the mean monthly RTEs up to the point that they were lost to follow-up and imputed the same value for all subsequent months of follow-up. We then compared mean cumulative RTEs between treatment arms using the paired t test.

We conducted sensitivity analyses to assess the impact of assumptions made in the study design. Sensitivity analyses examined 6- and 24-month follow-up periods and used RTEs derived from the latent class analysis in Phase 1 rather than the RTEs derived from the primary analysis of entire cohort in Phase 1. In the latent class analyses, we applied the RTEs for each latent class to all individuals rather than trying to assign an individual to a particular latent class. To assess the impact of imputation on our results, we repeated our primary analyses including only individuals with complete follow-up. To assess if our results were influenced by the choice of anti-TNF drug, we repeated analyses examining only those who received infliximab as their first anti-TNF. Additional sensitivity analyses assessed the impact of differences in the designs of the cohort study and the Markov model (described below). In 1 sensitivity analysis, we excluded those individuals entering follow-up with an ostomy. In another, we excluded those patients with ostomy at baseline and did not apply a reduction in quality of life for the use of antidiarrheal medications.

Simulation Model Construction and Analysis

We constructed a Markov model comparing the 2 treatment strategies to each other, again using an initial treatment-carried-forward approach to assign the treatment arm even if the simulated patient switched therapy during follow-up. We built the simulation to mirror the changes in medications that occurred in the cohort and used these changes in therapy to define subsequent health states; we also used the rates of rare adverse events from the cohort study to model these in the simulation. We based states within the model on current medication exposure, including infliximab, infliximab + CS, adalimumab, adalimumab + CS, certolizumab, certolizumab + CS, CS only, and no medical therapy with anti-TNF or CS drugs (Figure 3.1). The cycle length for the model was 1 month, and the time horizon for the base case was 1 year. We derived transition probabilities between states directly from the Phase 2 cohort study (see Supplemental Table 3.2). Adverse events in the model included all-cause mortality, bowel resection surgery, nondermatologic malignancies, acute myocardial infarction, stroke, acute infection, IBD-related hospitalization, and hip fracture. We derived rates of these adverse events from the Phase 2 results. We validated the model by comparing medication use as estimated by the model with actual use observed within the retrospective cohort at the end of 1 year, to ensure that the model adequately simulated changes in disease status appreciated in the retrospective cohort. We assigned rewards to each health state using RTEs as the common metric in the same manner as in the preference-weighted cohort study.

Figure 3.1

Overview of Markov Model Structure.

We calculated cumulative mean RTEs and 95% CIs within the model via First Order Monte Carlo simulation of 1000 trials of 9780 individuals, based on the original size of the cohort. We calculated incremental effectiveness by subtracting the mean values for each treatment strategy (anti-TNF and prolonged CS). We also compared these against the estimates derived from the retrospective cohort.

Secondary analyses examined the impact of different patient preferences by using different RTEs for health states. To do so, we repeated the models using RTEs derived from the latent class analysis in Phase 1.

We conducted sensitivity analyses to assess the impact of assumptions built into the model. We created 1-way sensitivity analyses and tornado plots for all transition probabilities and utility estimates within the Markov model to determine if either the retrospective analysis or simulation were sensitive to a specific parameter estimate. Another sensitivity analysis examined the way in which rewards (RTEs) were assigned to health states that were a priori deemed to last longer than 1 month, such as active disease at the time of initiation of therapy. In the primary analysis of the Markov model, we applied the rewards at the start of the first month in the health state. We repeated the analysis, applying estimates in a per-month fashion. In this iteration, we gave rewards over 2 months as opposed to at the start of the initial month of that state. A sensitivity analysis applied an additional reduction in quality of life for all months that patients were treated with CS, to capture the negative impact of this therapy as suggested from the results of Phase 1.

We performed an additional analysis that examined the impact of emergent and elective surgery and surgical outcomes on our results in the simulation model. The primary model assumed that all surgical events were elective and did not result in an ostomy. However, to assess the impact of surgical procedures being either emergent or elective, as well as the potential for both ostomy and nonostomy outcomes, we constructed a model that included these surgical events. We derived transition probabilities for these events from the retrospective cohort. We then compared these results against the base case analyses.

We built the Markov model using TreeAge Pro 2016 (TreeAge Software, Inc). We conducted all other analyses using SAS version 9.4 (SAS Institute).

Deviation From the Research Proposal Submitted to PCORI

In the application to PCORI, the investigators proposed that the model for aim 3 would use a discrete event simulation model. However, once aim 2 was complete, it was apparent that the data structure, which updated incidence rates and hazard ratios in a monthly fashion, was more naturally represented by a Markov model with 1-month cycle length as opposed to continuous event rates utilized in discrete event simulation. Therefore, we used a traditional Markov model.

Ethical Considerations and Patient Involvement

The study protocol was approved by the IRBs at University of Pennsylvania and University of Alabama at Birmingham. The study question was proposed and the study was designed—including the choice of outcome measures—and implemented by the investigators. Patient stakeholders provided feedback on the study design and results.

Results

Phase 1—Discrete Choice Experiment

In total, 1753 Crohn's disease patients were invited to participate. Of those, 81% (1422/1753) agreed to learn more about the study; 1409 were sent the consent page; and 814 responded, but 2 did not complete all of the choice questions. Thus, 58% (812/1409) of those who were sent the consent form completed the full choice-experiment survey (Figure 1.3). Five respondents failed the test for internal validity. Given the low number, they were not excluded from final analysis.

Figure 1.3

Participant Accruement for Phase 1.

The majority of respondents were female, consistent with the CCFA Partners' population (Table 1.1).¹³ More than a third of respondents reported having a complication with their Crohn's disease (abscess, stricture, or fistula) and the majority were not in a self-reported remission, although the median short Crohn's Disease Activity Index (SCDAI) score was 142. Approximately a third of patients were currently using an immunosuppressant medication and a third had used an oral corticosteroid in the prior year.

Table 1.1

Baseline Demographics for Phase 1.

Scaled Preference Utilities

We first assessed to make sure that the participants' preferences for the levels of each attribute included in the scenarios were logically ordered. A 1-class choice model produced average preference-utility estimates for the study population. Preference weights for categorical attribute levels are shown in Figure 1.4, with the relative importance of each attribute indicated by the distance or length of a vertical line between 0 and the greatest or most severe level of each attribute. This measure of relative importance depends on the range of levels evaluated. We evaluated all severity levels and steroid use over 12 months, but infection risk ranged as high as 30%, while the maximum value of cancer and surgery was only 8%. Preference weights showed logically ordered utility losses with increasing levels of risk and/or disease activity. Differences among symptom severity and risk levels generally were statistically significant. An exception was mild disease durations, where respondents were insensitive to differences in durations greater than 0; this risk-preference insensitivity has been previously shown in IBD patients.²³

Figure 1.4

Categorical Preference Weights.

Severe disease duration was the most important attribute, with 8% risk of surgery and mild disease duration being relatively unimportant (Figure 1.4). Four months of severe disease was 1.6 times and 3.2 times more important than 4 months of moderate or mild disease, respectively. The differences in importance were larger if the duration of active disease was longer. Twelve months of severe disease was 2.1 times more important than 12 months of moderate disease and 6.1 times more important than 12 months of mild disease, indicative of an interaction between duration and disease severity. At the 5% risk level, cancer risk was about twice as important as surgery risk and about 6 times more important than infection risk. Avoiding 12 months of steroid use, with perceived side effects, was more important than avoiding an 8% surgery risk.

Stratified Analysis Based on Age and Use of Immunosuppressant Medications

Relative importance for attributes was generally similar when comparing participants who were older than or younger than age 60 and for those with and without current immunosuppressant use (Figure 1.5). Comparing the relative importance of attributes among those younger and older than 60 years, the older population placed slightly more importance on avoiding risks. This was statistically significant only for the risk of surgery (P = .009 without adjusting for multiple comparisons).

Figure 1.5

Categorical Preference Weights Stratified by Age (A) and Use of Immunosuppressant Medications, (B) Including Thiopurine Analogs, Biologics, and Methotrexate.

Latent Class Analysis of Choice Data

We used latent class analysis to assess whether there were multiple different patterns in which patients have preferences for CD treatments. The best model fit was a 3-class model. Figure 1.6 compares the relative importance of the attributes by latent class. An “efficacy-seeking class” constituted 61% of the overall sample and had preferences similar to the overall sample, with a strong preference for efficacy, specifically avoiding moderate and severe disease. The “steroid-avoidant” class was 25% of our sample and had a strong preference for avoiding CS, even at cost of medication efficacy. The “risk-avoidant class” constituted 14% of the overall sample and preferred avoiding therapeutic risks, especially cancer risks.

Figure 1.6

Relative Attribute Importance by Latent Class*.

We performed an analysis to determine the number of months of remission (or “healthy time”) that patients in each of the latent classes were willing to forego to avoid various attribute levels (Figure 1.7 A through D and Supplemental Table 1.1). For example, for the risk-avoidant class, the effects-coded parameter estimates for no cancer risk and 5% cancer risk were 1.67 and −0.60, respectively. The mean marginal remission time utility was about 0.21. Thus, the risk-avoidant class RTE for having to bear a cancer-risk exposure of 5% was calculated using Equation 1 as

Figure 1.7

Remission Time Equivalents by Latent Class*.

$$RRRRRR\left(r{r}_{cccccccccccc}=0.05|cccccccccc=rrrrccrrccrrrrrrrr\right)=\frac{-0.60-1.67}{0.21}=-10.7mmrrmmmmhcc.$$

The RTEs for the class most concerned with medication efficacy most closely approximated the RTEs for the overall sample. However, in each level of attributes, the mean RTEs for the overall sample underestimated the RTEs for at least 1 of the groups.

The heterogeneity in treatment preferences results in markedly different valuation of durations of active Crohn's disease (Table 1.2). For example, the efficacy-seeking class valued 3 months of moderate disease (−4.1 RTEs; 95% CI, −5.3 to −2.9) similarly as the steroid-avoidant class valued 5 months of moderate disease (−3.8 RTEs; 95% CI, −6.4 to −1.2) and the risk-avoidant class valued 9 months of moderate disease (−3.9 RTEs; 95% CI, −8.1 to 0.2). Similarly, the efficacy-seeking class valued 3 months of moderate disease with greater disutility (−4.1 RTEs) compared with those in the steroid-avoidant class (−2.5 RTEs) or the risk-avoidant class (−2.5 RTEs).

Table 1.2

Remission Time Equivalents.

Beta coefficients for each attribute derived from the 1-class and 3-class models are provided in Supplemental Table 1.2.

Latent Class Membership

We evaluated what covariates predicted increased likelihood of membership in each of the 3 latent classes. Clinically relevant covariates were added and removed sequentially based on significance at a P value of ≤.05. Four covariates retained significance in the model (Table 1.3). Overall, the corticosteroid-avoidant class tended to be female and older, whereas the efficacy-seeking participants were more often younger and with lower SCDAI scores.

Table 1.3

Participant Characteristics by Latent Class Membership.

Phase 2—Comparative Effectiveness Study

Among patients with CD, 7694 entered the cohort as prolonged CS users and 1879 as new anti-TNF users. Among patients with UC, 3224 and 459 entered the cohort as prolonged CS users and new anti-TNF users, respectively (Figure 2.1). After the start of follow-up, 1473 patients with CD and 462 patients with UC initiated anti-TNF such that overall there were 3352 patients with CD and 921 with UC who initiated anti-TNF therapy during follow-up (Figure 2.1). The total duration of follow-up time (in person-years [median, interquartile range]) was 19 667 (1.8, 0.7-4.0) for CD treated with prolonged CS, 11 180 (2.6, 1.1-4.9) for CD treated with anti-TNF, 8121 (1.9, 0.8-4.1) for UC treated with prolonged CS, and 2250 (1.9, 0.8-3.9) for UC treated with anti-TNF. The weighted baseline characteristics were generally well balanced between the treatment groups (Tables 2.1 and 2.2 and Supplemental Tables 2.1 and 2.2). The mean stabilized weights remained similar between treatment groups throughout the follow-up period (Supplemental Figure 2.1). Use of CS during the 12 months before the start of follow-up was greater in the patients who entered the cohort as prolonged CS users (P < .01 for all comparisons). CS use decreased significantly in the anti-TNF group during the first 6 months and in both groups of patients during the first year of follow-up (Table 2.3).

Figure 2.1

Creation of Study Cohorts.

Table 2.1

Characteristics of the Crohn's Disease Study Population in the Retrospective Cohort Study at the Start of Follow-up.

Table 2.2

Characteristics of the Ulcerative Colitis Study Population in the Retrospective Cohort Study at the Start of Follow-up.

Table 2.3

Steroid Use in the 12 Months Before Start of Follow-up and in the First Year of Follow-up.

Primary Outcome

The weighted annual incidence of death per 1000 treated CD patients was 21.4 (anti-TNF) vs 30.1 (prolonged CS); in UC patients, the incidence of death was 23.0 (anti-TNF) and 30.9 (prolonged CS; Figure 2.2). The risk of death was statistically significantly lower in patients treated with anti-TNF therapy for CD (odds ratio [OR], 0.78; 95% CI, 0.65-0.93) and numerically lower but not significantly different for UC (OR, 0.87; 95% CI, 0.63-1.22). The results were similar in sensitivity analyses modifying the exposure definition (Figure 2.3). Of note, in unweighted models not accounting for confounding, the association between treatment and mortality were generally stronger in the primary analyses (CD OR, 0.65; 95% CI, 0.57-0.75; UC OR, 0.77; 95% CI, 0.59-1.01) and reached statistical significance in most of the sensitivity analyses (Supplemental Table 2.3).

Figure 2.2

Adjusted Odds Ratios for Primary and Secondary Outcomes in Crohn's Disease (A) and Ulcerative Colitis (B).

Figure 2.3

Sensitivity Analysis Examining Different Definitions of Exposure.

The association between anti-TNF therapy and mortality did not vary by age (Table 2.4; interaction P > .6 for both diseases). In contrast, when stratified by comorbid illness, anti-TNF therapy was associated with a reduced mortality risk only among patients who had the most comorbid illnesses (CD OR, 0.65; 95% CI, 0.48-0.88; UC OR, 0.63; 95% CI, 0.36-1.11), albeit the test for interaction was not statistically significant (CD P = .17; UC P = .29).

Table 2.4

Stratified Analysis of the Association of Anti-TNF Therapy Relative to CS Therapy and the Risk of Death Among Patients With IBD.

Secondary Outcomes

Among the CD cohort, anti-TNF therapy was also associated with lower rates of MACE (OR, 0.68; 95% CI, 0.55-0.85) and hip fracture (OR, 0.54; 95% CI, 0.34-0.83; Figure 2.2). The risk of serious infection, pulmonary embolus, and cancer was not significantly different between treatments for CD, although the risk of cancer nearly met statistical significance (OR, 1.27; 95% CI, 0.98-1.65). None of these secondary outcomes were significantly different between treatments for UC.

The difference in emergency bowel resection was not statistically significant for CD (OR, 1.17; 95% CI, 0.96-1.42). For UC, emergency surgery was more common in patients treated with anti-TNF therapy (OR, 2.18; 95% CI, 1.37-3.46). IBD-related hospitalizations were slightly more common in the anti-TNF treated group (CD OR, 1.13; 95% CI, 1.04-1.23; UC OR, 1.53; 95% CI, 1.29-1.81). The risk of serious infection was not statistically significant for CD (OR, 0.98; 95% CI, 0.87-1.10) or UC (OR, 0.99; 95% CI, 0.78-1.26).

To determine whether the increased risk of death with CS therapy among patients with CD could be explained based on the measured secondary outcomes, we repeated the primary analysis censoring follow-up at the time of any of the secondary outcomes. In this model, the reduced risk for death was attenuated and very close to a null result (OR, 0.97; 95% CI, 0.63-1.47).

Phase 3—Preference-Weighted Cohort Study

In the retrospective cohort study of patients with CD, 1879 patients categorized as anti-TNF users and 7694 patients categorized as CS users were identified after applying inclusion and exclusion criteria. After removing outliers via propensity score analysis, 1563 patients with CD identified as new anti-TNF initiators were matched 1:1 to 1563 individuals identified as prolonged CS utilizers. After propensity score matching, all baseline covariates had a SMD ≤0.1 (Supplemental Table 3.1). Only 14.7% of the prolonged CS cohort and 11.5% of the anti-TNF cohort were lost to follow-up before 1 year. Cumulative risks of each adverse event for both anti-TNF users and the prolonged CS users are presented in Table 3.2. Transitions between medications during the year are described in Supplemental Tables 3.2a through 3.2d.

Table 3.2

Cumulative Adverse Events for Both Anti-TNF Users and the Prolonged CS Users.

In this cohort, anti-TNF therapy yielded more RTEs than a strategy of continued intermittent CS use (Table 3.3). This benefit was evident at 6 months (mean incremental difference [MID], 0.5; 95% CI, 0.4-0.6), 12 months (MID, 0.8 RTEs; 95% CI, 0.5-1.1), and 24 months (MID, 1.4; 95% CI, 0.9-2.0). Restricting our results to only those with complete follow-up slightly reduced our cohort size but did not markedly affect the appreciated benefit of initiating anti-TNF medications (Table 3.3). When assessing only those who used infliximab as their initial anti-TNF, the estimates were similar to those appreciated in our primary analysis at 6, 12, and 24 months. Restricting the analyses to only those without an ostomy at baseline and excluding the impact of concomitant antidiarrheal medications also did not affect the preferred strategy. In latent class analyses, anti-TNFs remained the preferred therapeutic strategy for each group, with the greatest potential RTE benefit appreciated by the efficacy-seeking group, and the least by the steroid-avoidant class (Table 3.4).

Table 3.3

Summary of Primary and Sensitivity Analyses From the Retrospective Cohort Study.

Table 3.4

Results of Subgroup Analyses Based on Latent Classes Within Retrospective Cohort Study.

Simulation Model

We used a Markov simulation model to further test the difference between the 2 treatment strategies, including assessing a wide range of sensitivity analyses. Transition rates and final states at the end of 1 year closely mirrored those in the retrospective cohort at the end of 1 year, demonstrating the internal validity of the simulation model (Table 3.5). The simulation model also estimated that anti-TNF initiation also produced a greater number of RTEs when compared with prolonged CS use, although the incremental difference determined by the model was smaller (Table 3.6). When incorporating RTE estimates for specific latent classes, both the efficacy-seeking class and the risk-avoidant class demonstrated that anti-TNFs were the preferred strategy. Surprisingly, the preferred strategy for the steroid-avoidant class was prolonged steroid use. However, when incorporating RTE estimates that reflected the strong disinterest of continued steroid use, anti-TNF use again became the preferred strategy in the steroid-avoidant class. Anti-TNFs also remained preferred in all other latent classes when incorporating these preferences.

Table 3.5

Comparison of Medication Utilization States at the End of 1 Year in Retrospective Cohort and Simulation Model.

Table 3.6

Comparison of Expected RTEs Over 12 Months With Anti-TNF Therapy or Continued Intermittent Use of Steroids Within the Markov Simulation Model.

The simulation model was not sensitive to 25% variation of any transition probability (Figure 3.2) nor was it sensitive to 15% variation of remission time equivalent estimates (Figure 3.3). Incorporation of emergent surgical events and postoperative ostomy states did not affect the preferred strategy, with anti-TNF use remaining preferred to prolonged CS use.

Figure 3.2

One-Way Sensitivity Analyses of Transition Probability Estimates in the Markov Simulation Model.

Figure 3.3

One-Way Sensitivity Analyses of RTE Estimates in the Markov Simulation Model.

Discussion

Patients with IBD and their physicians struggle to choose between intermittent immunosuppression with CS and chronic immunosuppression with anti-TNF therapy. Important factors in these decisions are the relative efficacy, toxicity, and anticipated impact on the patient's quality of life. The latter can be influenced by disease activity, the occurrence of adverse events, or even the fear of adverse events. The decisions are even more complicated for elderly patients and those with significant comorbidities, where the risks of complications may be higher and the consequences more severe. This study was designed to inform this complicated treatment decision, using state-of-the-art methods that include discrete choice experiment to derive RTEs for CD, a cohort study employing marginal structural models, a preference-weighted cohort study using PS matching and the RTE weights derived from Phase 1, and a Markov simulation to test the robustness of the results of the preference-weighted cohort study.

Combining these results, we were able to demonstrate at least 3 distinct patterns by which patients with CD make decisions about medical treatments, that the choice to continue using CS as the primary therapy for CD is associated with higher mortality rates than treatment with anti-TNF drugs as steroid-sparing agents, and that quality of life is also better in those patients treated with anti-TNF drugs. These results strongly support a strategy of using anti-TNF drugs as steroid-sparing therapy in patients with CD who require a second or prolonged steroid course within a 12-month period. Among similar patients with UC, the observed reduction in mortality was not statistically significant, but the results were generally in the same direction as those for CD. Additional modeling and preference-weighted cohort studies will be needed to assess whether predicted quality of life is improved with anti-TNF therapy relative to CS for UC.

Each phase of this work has made important contributions to our knowledge of preferred treatment options and/or the methods of clinical research. In Phase 1, we demonstrated substantial preference heterogeneity among patients with CD. We identified 3 subgroups within our sample who valued treatment options and health states differently. The largest group emphasized minimizing the time spent with moderate to severe disease activity. Members of this class were less concerned with corticosteroid use or the risks associated with medication-related adverse effects or surgery for their CD. In contrast, 39% of the patients were best represented by 1 of 2 other latent classes—1 group was concerned about corticosteroid use and 1 was most concerned about the risks of medication-related adverse events. For example, the risk-avoidant class was willing to accept nearly 1 year of severely active disease to avoid a 5% risk of cancer. The corticosteroid-avoidant class was willing to accept nearly 4 months of moderately active disease to avoid 2 months of corticosteroid exposure. These data demonstrate that health policies and guidelines that consider all patients with Crohn's disease as a single group may fail to meet the needs of up to a third of the population whose preferences are significantly different from the majority.

Phase 1 also advanced the field by translating patient preferences into RTEs as opposed to utilities that are bound between 0 and 1 and assume a linear relationship to define quality-adjusted life years. The RTEs derived from this discrete choice experiment directly calculate the value of severity durations. They do not require calculating quality-adjusted time by multiplying durations by a cardinal measure of utility between 0 and 1. RTEs thus allow for nonlinearities in severity-duration and risk-bearing preferences for clinically relevant outcomes and time periods. This was more evident for the duration of mild symptoms than for moderate or severe CD symptoms where the relationship with time was more linear. RTEs also are not limited to the 12-month maximum treatment period. For example, 12 months of severe disease logically was equivalent to losing more than 19 months of remission. This does not necessarily imply that this would be worse than death, but it is consistent with the severe disutility that would be experienced from prolonged periods of severely active disease, a condition for which almost all patients would undergo some major intervention. While computations using health state utilities and quality-adjusted life years are difficult to interpret for relatively short, clinically relevant durations of ill health, RTEs provide an alternative metric that is conceptually consistent with the outcomes of interest and may be easier to understand by patients, providers, and policy makers. However, the RTEs are disease specific and, as such, are not easily compared across disease states.

Following completion of the study, we presented our results individually to our patient and educator stakeholder advisors. Some of the stakeholders emphasized that seeing the results helped them consider how they make their own decisions and that knowledge of this could influence how they make decisions in the future.

There are potential limitations to Phase 1. Choice experiments involve a hypothetical decision-making experience and patients could make different choices in real life because of the quite different clinical, financial, and emotional consequences of actual treatment decisions. To minimize potential bias, scenarios were presented as realistically as possible and emphasized the importance of the research and the need for full concentration when answering the questions. The number of questions that a participant needed to answer was also limited to avoid mental fatigue, since choice experiments are cognitively challenging. Internal validity testing demonstrated good understanding of the choice tasks.

The treatment trait for time needing to be treated with CS may have been interpreted differently by each participant based on his or her personal experience with CS, whereas the other traits focused on specific adverse events that could result from immunosuppression therapy. We could have included additional traits to describe potential therapies, such as the need for intravenous or subcutaneous administration; the risk of other complications such as hip fracture or cardiac disease, which proved important in Phase 2; and so on. It is unlikely that these would have meaningfully affected the calculations of RTEs or the 3 latent classes, but they could impact the way patients actually choose between treatments in real-world settings. Due to potential confusion over conditional probabilities, outcomes and risks were presented as certain. Some of the RTE estimates had wide confidence intervals due to relatively few participants with choice patterns consistent with a specific latent class group answering questions that contribute to these estimates. Finally, while informative about patterns of preference distributions, latent class analysis is probabilistic and does not predict with certainty that a given patient or patient group will fall into a specific preference class. Nevertheless, physicians are likely to find the 3 statistical classes we identified as corresponding to actual types of patients they encounter clinically.

In Phase 2, we used state-of-the-art marginal structural models with stabilized weights from inverse probability of treatment derived from propensity score models and models estimating the probability of being censored⁶⁶ to estimate the comparative effectiveness of CS and anti-TNF drugs, the 2 most commonly used treatment strategies to control active symptoms of CD and UC. We selected death as our primary outcome because reluctance to use chronic immunosuppression is usually related to fear of serious adverse events, particularly in the elderly or in patients with comorbid illnesses that further increase the risk of death.^76,77 Patients who were treated with anti-TNF therapy for CD had a lower risk of death, which was most evident in those individuals at greatest risk (ie, those with the most comorbid illnesses). As demonstrated by our model censoring for any of the secondary outcomes, essentially all of the increased risk of death associated with CS therapy could be explained by the excess risk of MACE, pulmonary embolus, hip fracture, serious infection, cancer, and emergency surgery, although treatment with CS was statistically significantly associated only with higher risk of MACE and hip fracture. The risk of death was numerically but not significantly lower in the UC patients treated with TNF therapy.

The results of this study are consistent with placebo-controlled trials of anti-TNF therapy where there was no increased risk of short-term mortality.⁷⁸ However, clinical trial populations do not reflect the broader use of therapies in clinical practice, often excluding patients with multiple comorbid illnesses and the elderly. Observational data can extend observations of clinical trials to populations who were excluded. Notably, the reduced mortality observed with anti-TNF therapy relative to CS therapy in this study was largely evident in those with the greatest burden of comorbid illness, while those with less severe or fewer comorbidities had comparable survival regardless of the choice of therapy. A small study from Belgium also observed that among elderly patients with IBD, those treated with CS were more likely to die than those treated with anti-TNF therapy.⁷⁷ These results, and those of a recent study examining outcomes of surgical vs medical therapy for UC,⁶⁴ suggest that episodic CS may be the least favorable strategy for this vulnerable population.

This study examined many the common causes of death as secondary outcomes. Some of the observations were expected. Hospitalization and emergency surgery were more common in anti-TNF treated patients, particularly in UC, likely because anti-TNF therapy was attempted in patients who were likely to require intravenous CS or surgery if the anti-TNF therapy was unsuccessful. Anti-TNF is considered to be one of the most effective therapies for ulcerative colitis and when ineffective in controlling symptoms, intravenous CS, administered as an inpatient, is often attempted as salvage therapy. When this fails, surgery is often required. Thus, the observed associations with hospitalization and emergency surgery are not unexpected and likely represent residual confounding by indication.

Osteoporosis is a known complication of prolonged CS therapy. Unsurprisingly, higher hip fracture rates were evident with CS treatment, particularly among patients with CD. This highlights the need to minimize CS exposure and assess for bone loss in patients who received prolonged CS therapy. Among patients with CD, anti-TNF therapy was associated with lower rates of MACE. However, the same effect was not observed among patients with UC. There are limited data on the effect of anti-TNF therapy on the risk of acute cardiovascular events among patients with IBD. However, some, but not all, studies in rheumatoid arthritis suggest that anti-TNF therapy may reduce the risk of myocardial infarction,⁷⁹ particularly if the inflammation is well controlled.⁸⁰ Similar studies examining the risk of cardiovascular outcomes with CS have produced conflicting results, with a suggestion that inflammatory disease activity may be a more important risk factor than CS use.⁸¹ Because of the nature of the 2 diseases, systemic inflammation is more common in CD than in UC.⁸² Whether this explains differences in associations of anti-TNF therapy with MACE between the 2 diseases is unknown.

Several unique aspects of the methods used in Phase 2 are noteworthy. Marginal structural models are superior to standard Cox models to account for confounding by time-updating factors that are associated with treatment selection and the outcome of interest.⁶⁶ The sample size was large, particularly for CD, allowing us to study relatively uncommon outcomes and to stratify results based on extent of comorbid illness. Because patients rarely discontinue Medicare insurance, loss to follow-up was minimized. Medicare covers inpatient and outpatient care and that from specialists and primary care physicians, thereby allowing capture of events that may be missed by examining only records of the treating gastroenterologists. Follow-up time was relatively long, with 25% of the cohort having more than 4 years of follow-up. Sensitivity analysis showed that the association of treatment strategy with mortality was robust to multiple assumptions and comparable during the first year and with longer duration of follow-up, suggesting that the relative risk does not escalate further with longer-term exposure. Unlike the TREAT registry, this study focused on patients who were initiating therapy with anti-TNF drugs rather than prevalent users, thereby avoiding the potential for bias from depletion of susceptible subjects. Finally, the study documents that steroid-sparing therapies are often not employed in usual care despite current treatment guidelines.^83,84

However, the study had several limitations. Despite adjusting for a large range of covariates in the marginal structural models, residual confounding is possible. Specifically, it was not possible to adjust for smoking, which is associated with all-cause mortality and nearly all of the major causes of death included as secondary outcomes in this study, as smoking is not reliably captured in these data. The risk of death among current smokers is estimated to be 1.5- to 2-fold greater than in nonsmokers.^85,86 Smoking is also associated with more severe CD, but recent studies have demonstrated that there is relatively little difference in the proportion of smokers and nonsmokers treated with CS or anti-TNF therapy for CD.^87,88 Thus, it is unlikely that residual confounding by smoking would fully explain the observed associations between treatment with CS vs anti-TNF and mortality. Some patients categorized as CD may have had UC and vice versa. Based on our results, this would likely have biased the significant association observed in CD toward the null and the nonsignificant findings for UC away from the null. Thus, any misclassification of this type is unlikely to have led to the wrong interpretation.

Phase 3 was perhaps the most novel of the 3 phases of this study, combining patient preferences as measured with RTEs with a longitudinal cohort study using health insurance claims data and creating a Markov simulation model to test the robustness of the findings. The Phase 3 data demonstrated that not only was anti-TNF therapy associated with a reduction in mortality, but patients with CD treated with anti-TNF therapy would be expected to have improved quality of life. The marginal benefit of anti-TNF therapy was approximately equivalent to 0.8 months of remission within the first year of therapy alone. An analysis using the preference patterns similar to the efficacy group from Phase 1 demonstrated even larger benefits: more than 1 month of remission within the first year of therapy. As noted previously, the RTE metric is easier to interpret than traditional quality-adjusted life years, as 1 month of remission does not depend on other aspects of the patient's health (ie, a patient with multiple sclerosis and a patient otherwise in perfect health can each interpret the value of 1 month of remission within his or her own framework).

Despite the highly statistically significant differences in quality of life among patients treated with CS or anti-TNF drugs, there may still be scenarios in which some individuals would have a better quality if treated with CS alone. In the discrete choice experiment that gave rise to the RTE estimates for Phase 3, patients had strongly negative preferences for treatments that increased the risk of cancer. In the overall population, a 2% increased risk of cancer in the first year of therapy was associated with a reduction of −1.5 RTEs, slightly larger than the net benefit of anti-TNF drugs relative to CS. For the efficacy-seeking class, the magnitude was much smaller (−0.3 RTEs), while among the risk-avoidant latent class the magnitude was much greater (−4.8 RTEs). Importantly, these computations are independent of what the true risk of cancer is. Rather, they represent the reduction in quality of life experienced by individuals who perceive that the medication they are taking increases risk of cancer by 2% per year relative to the alternative treatment. Most studies, including this one, suggest that anti-TNF therapy presents a very small, if any, increased risk of cancer other than nonmelanoma skin cancer.⁸⁹ As such, appropriately describing risks, preferably as absolute risk rather than relative risk, is essential to allow rational decision-making by physicians and patients.

As an analytic approach, modeling studies are potentially subject to drawing the wrong conclusions due to the quality of the data used to inform the model. Most models draw from clinical trials whenever possible to inform transition probabilities. However, patients in clinical trials may differ substantially from those in routine practice and different clinical trials may enroll very different patients. By drawing nearly all of our transition probabilities from the retrospective cohort study using advanced methods to account for confounding factors, we increased the likelihood that the transition probabilities were drawn from otherwise similar populations. In addition, we were able to internally validate our model against the Medicare and Medicaid cohort by examining the rates of outcomes and medication patterns over time. However, we still needed to make many assumptions, such as our definition of remission, which was based on expert opinion derived from treatment patterns observable in the claims data. Furthermore, we used our Markov model to allow for additional sensitivity analyses, but the model did not exactly mirror the cohort study. This reflects the challenge of trying to capture the complex nature of a disease course in a mathematical model.

Conclusions

CD and UC are chronic debilitating and potentially fatal diseases. The management of these diseases has made exponential advances over the past 2 decades, yet these data demonstrate that many patients continue to receive prolonged courses of CS despite the availability of new and more effective therapies. This study was not specifically designed to determine the underlying reasons for the observed treatment patterns. However, in Phase 1 we demonstrated that a subgroup of patients strongly emphasizes the fear of complications of medical therapy when choosing between therapeutic alternatives. Ironically, the anti-TNF medications, which are often more feared by patients and physicians because of the perceived risk of adverse events, were associated with a reduced risk of mortality when compared head-to-head with a strategy of continued intermittent courses of CS therapy for CD. Not surprisingly, much of the mortality effect was able to be explained by an increased risk of cardiovascular events and hip fractures in patients treated with CS, both of which are known complications of CS therapy. Moreover, our preference-weighted cohort study and simulation model demonstrated that quality of life is also better with anti-TNF treatment and that the results were robust when tested in a wide range of sensitivity analyses. Taken together, these data strongly emphasize that patients with CD, and perhaps UC, who are prescribed prolonged and repeated courses of CS to manage their symptoms would be better served to try a course of anti-TNF therapy as a steroid-sparing approach. Although this study did not address whether anti-TNF medications should be used before CS as a first-line therapy for CD, these data make a clinical trial addressing this question intriguing to consider.

Poststudy addendum: The investigators intended to conduct the analyses in Phase 1 and Phase 3 for UC as well but were unable to be complete them in the timeline allowed within the PCORI contract.

References

1.

Herrinton LJ, Liu L, Lafata JE, et al. Estimation of the period prevalence of inflammatory bowel disease among nine health plans using computerized diagnoses and outpatient pharmacy dispensings. Inflamm Bowel Dis. 2007;13(4):451-461. [PubMed: 17219403]

2.

Herrinton LJ, Liu L, Lewis JD, Griffin PM, Allison J. Incidence and prevalence of inflammatory bowel disease in a Northern California managed care organization, 1996-2002. Am J Gastroenterol. 2008;103(8):1998-2006. [PubMed: 18796097]

3.

Loftus EV Jr, Schoenfeld P, Sandborn WJ. The epidemiology and natural history of Crohn's disease in population-based patient cohorts from North America: a systematic review. Aliment Pharmacol Ther. 2002;16(1):51-60. [PubMed: 11856078]

4.

Loftus EV Jr, Silverstein MD, Sandborn WJ, Tremaine WJ, Harmsen WS, Zinsmeister AR. Crohn's disease in Olmsted County, Minnesota, 1940-1993: incidence, prevalence, and survival. Gastroenterology. 1998;114(6):1161-1168. [PubMed: 9609752]

5.

Loftus EV Jr, Silverstein MD, Sandborn WJ, Tremaine WJ, Harmsen WS, Zinsmeister AR. Ulcerative colitis in Olmsted County, Minnesota, 1940-1993: incidence, prevalence, and survival. Gut. 2000;46(3):336-343. [PMC free article: PMC1727835] [PubMed: 10673294]

6.

Burisch J, Jess T, Martinato M, Lakatos PL, EpiCom E. The burden of inflammatory bowel disease in Europe. J Crohns Colitis. 2013;7(4):322-337. [PubMed: 23395397]

7.

Canavan C, Abrams KR, Mayberry JF. Meta-analysis: mortality in Crohn's disease. Aliment Pharmacol Ther. 2007;25(8):861-870. [PubMed: 17402989]

8.

Lewis JD, Gelfand JM, Troxel AB, et al. Immunosuppressant medications and mortality in inflammatory bowel disease. Am J Gastroenterol. 2008;103(6):1428-1435, quiz 1436. [PubMed: 18494836]

9.

Ananthakrishnan AN, Weber LR, Knox JF, et al. Permanent work disability in Crohn's disease. Am J Gastroenterol. 2008;103(1):154-161. [PubMed: 18076736]

10.

Lichtenstein GR, Yan S, Bala M, Hanauer S. Remission in patients with Crohn's disease is associated with improvement in employment and quality of life and a decrease in hospitalizations and surgeries. Am J Gastroenterol. 2004;99(1):91-96. [PubMed: 14687148]

11.

Drossman DA, Patrick DL, Mitchell CM, Zagami EA, Appelbaum MI. Health-related quality of life in inflammatory bowel disease. functional status and patient worries and concerns. Dig Dis Sci. 1989;34(9):1379-1386. [PubMed: 2766905]

12.

Lichtenstein GR, Hanauer SB, Sandborn WJ. Management of Crohn's disease in adults. Am J Gastroenterol. 2009;104(2):465-483, quiz 464, 484. [PubMed: 19174807]

13.

Chande N, Marshall JK, Seow CH, et al. New applications for traditional drugs in inflammatory bowel disease: what do Cochrane Reviews tell us? Inflamm Bowel Dis. 2015;21(12):2948-2957. [PubMed: 26540276]

14.

Deepak P, Bruining DH. Update on the medical management of Crohn's disease. Curr Gastroenterol Rep. 2015;17(11):41. [PubMed: 26363802]

15.

Affronti A, Orlando A, Cottone M. An update on medical management on Crohn's disease. Expert Opin Pharmacother. 2015;16(1):63-78. [PubMed: 25418125]

16.

Sandborn WJ, Feagan BG, Lichtenstein GR. Medical management of mild to moderate Crohn's disease: evidence-based treatment algorithms for induction and maintenance of remission. Aliment Pharmacol Ther. 2007;26(7):987-1003. [PubMed: 17877506]

17.

D'Haens G, Baert F, van Assche G, et al. Early combined immunosuppression or conventional management in patients with newly diagnosed Crohn's disease: an open randomised trial. Lancet. 2008;371(9613):660-667. [PubMed: 18295023]

18.

Faubion WA Jr, Loftus EV Jr, Harmsen WS, Zinsmeister AR, Sandborn WJ. The natural history of corticosteroid therapy for inflammatory bowel disease: a population-based study. Gastroenterology. 2001;121(2):255-260. [PubMed: 11487534]

19.

Hanauer SB, Stromberg U. Oral Pentasa in the treatment of active Crohn's disease: a meta-analysis of double-blind, placebo-controlled trials. Clin Gastroenterol Hepatol. 2004;2(5):379-388. [PubMed: 15118975]

20.

Sutherland L, Macdonald JK. Oral 5-aminosalicylic acid for maintenance of remission in ulcerative colitis. Cochrane Database Syst Rev. 2006;(2):CD000544. doi:10.1002/14651858.CD000544.pub4. [PubMed: 16625537] [CrossRef]

21.

Sutherland L, Macdonald JK. Oral 5-aminosalicylic acid for induction of remission in ulcerative colitis. Cochrane Database Syst Rev. 2006;(2):CD000543. doi:10.1002/14651858.CD000543. [PubMed: 16625536] [CrossRef]

22.

Sandborn WJ, Rutgeerts P, Feagan BG, et al. Colectomy rate comparison after treatment of ulcerative colitis with placebo or infliximab. Gastroenterology. 2009;137(4):1250-1260, quiz 1520. [PubMed: 19596014]

23.

Lichtenstein GR, Yan S, Bala M, Blank M, Sands BE. Infliximab maintenance treatment reduces hospitalizations, surgeries, and procedures in fistulizing Crohn's disease. Gastroenterology. 2005;128(4):862-869. [PubMed: 15825070]

24.

Rutgeerts P, Feagan BG, Lichtenstein GR, et al. Comparison of scheduled and episodic treatment strategies of infliximab in Crohn's disease. Gastroenterology. 2004;126(2):402-413. [PubMed: 14762776]

25.

Feagan BG, Panaccione R, Sandborn WJ, et al. Effects of adalimumab therapy on incidence of hospitalization and surgery in Crohn's disease: results from the CHARM study. Gastroenterology. 2008;135(5):1493-1499. [PubMed: 18848553]

26.

Colombel JF, Sandborn WJ, Reinisch W, et al. Infliximab, azathioprine, or combination therapy for Crohn's disease. N Engl J Med. 2010;362(15):1383-1395. [PubMed: 20393175]

27.

Colombel JF, Sandborn WJ, Rutgeerts P, et al. Adalimumab for maintenance of clinical response and remission in patients with Crohn's disease: the CHARM trial. Gastroenterology. 2007;132(1):52-65. [PubMed: 17241859]

28.

Panaccione R, Ghosh S, Middleton S, et al. Combination therapy with infliximab and azathioprine is superior to monotherapy with either agent in ulcerative colitis. Gastroenterology 2014;146(2):392-400 e393. [PubMed: 24512909]

29.

Johnson FR, Ozdemir S, Mansfield C, et al. Crohn's disease patients' risk-benefit preferences: serious adverse event risks versus treatment efficacy. Gastroenterology. 2007;133(3):769-779. [PubMed: 17628557]

30.

Peppercom MA. 6-mercaptopurine for the management of ulcerative colitis: a concept whose time has come. Am J Gastroenterol. 1996;91(9):1689-1690. [PubMed: 8792681]

31.

Sands BE, Siegel C, Ozdeir S. Gastroenterologists' tolerance for Crohn's disease treatment risks. Am J Gastroenterol. 2007;102(Suppl 2):S492.

32.

Grijalva CG, Chen L, Delzell E, et al. Initiation of tumor necrosis factor-alpha antagonists and the risk of hospitalization for infection in patients with autoimmune diseases. JAMA. 2011;306(21):2331-2339. [PMC free article: PMC3428224] [PubMed: 22056398]

33.

Gupta G, Lautenbach E, Lewis JD. Incidence and risk factors for herpes zoster among patients with inflammatory bowel disease. Clin Gastroenterol Hepatol. 2006;4(12):1483-1490. [PubMed: 17162240]

34.

Toruner M, Loftus EV Jr, Harmsen WS, et al. Risk factors for opportunistic infections in patients with inflammatory bowel disease. Gastroenterology. 2008;134(4):929-936. [PubMed: 18294633]

35.

Lichtenstein GR, Feagan BG, Cohen RD, et al. Serious infections and mortality in association with therapies for Crohn's disease: TREAT registry. Clin Gastroenterol Hepatol. 2006;4(5):621-630. [PubMed: 16678077]

36.

Long MD, Farraye FA, Okafor PN, Martin C, Sandler RS, Kappelman MD. Increased risk of pneumocystis jiroveci pneumonia among patients with inflammatory bowel disease. Inflamm Bowel Dis. 2013;19(5):1018-1024. [PMC free article: PMC3879785] [PubMed: 23478805]

37.

Long MD, Martin C, Sandler RS, Kappelman MD. Increased risk of pneumonia among patients with inflammatory bowel disease. Am J Gastroenterol. 2013;108(2):240-248. [PMC free article: PMC4624299] [PubMed: 23295276]

38.

Chung ES, Packer M, Lo KH, Fasanmade AA, Willerson JT, Anti-TNF Therapy Against Congestive Heart Failure Investigators. Randomized, double-blind, placebo-controlled, pilot trial of infliximab, a chimeric monoclonal antibody to tumor necrosis factor-alpha, in patients with moderate-to-severe heart failure: results of the Anti-TNF Therapy Against Congestive Heart Failure (ATTACH) trial. Circulation. 2003;107(25):3133-3140. [PubMed: 12796126]

39.

Feenstra J, Grobbee DE, Remme WJ, Stricker BH. Drug-induced heart failure. J Am Coll Cardiol. 1999;33(5):1152-1162. [PubMed: 10193711]

40.

Beaugerie L, Brousse N, Bouvier AM, et al. Lymphoproliferative disorders in patients receiving thiopurines for inflammatory bowel disease: a prospective observational cohort study. Lancet. 2009;374(9701):1617-1625. [PubMed: 19837455]

41.

Siegel CA, Marden SM, Persing SM, Larson RJ, Sands BE. Risk of lymphoma associated with combination anti-tumor necrosis factor and immunomodulator therapy for the treatment of Crohn's disease: a meta-analysis. Clin Gastroenterol Hepatol. 2009;7(8):874-881. [PMC free article: PMC2846413] [PubMed: 19558997]

42.

Bernstein CN, Blanchard JF, Metge C, Yogendran M. The association between corticosteroid use and development of fractures among IBD patients in a population-based database. Am J Gastroenterol. 2003;98(8):1797-1801. [PubMed: 12907335]

43.

van Staa TP, Cooper C, Brusse LS, Leufkens H, Javaid MK, Arden NK. Inflammatory bowel disease and the risk of fracture. Gastroenterology. 2003;125(6):1591-1597. [PubMed: 14724810]

44.

Higgins PD, Skup M, Mulani PM, Lin J, Chao J. Increased risk of venous thromboembolic events with corticosteroid vs biologic therapy for inflammatory bowel disease. Clin Gastroenterol Hepatol. 2015;13(2):316-321. [PubMed: 25038374]

45.

Lichtenstein GR, Feagan BG, Cohen RD, et al. Serious infection and mortality in patients with Crohn's disease: more than 5 years of follow-up in the TREAT registry. Am J Gastroenterol. 2012;107(9):1409-1422. [PMC free article: PMC3438468] [PubMed: 22890223]

46.

Khanna R, Bressler B, Levesque BG, et al. Early combined immunosuppression for the management of Crohn's disease (REACT): a cluster randomised controlled trial. Lancet. 2015;386(10006):1825-1834. [PubMed: 26342731]

47.

Fidder H, Schnitzler F, Ferrante M, et al. Long-term safety of infliximab for the treatment of inflammatory bowel disease: a single-centre cohort study. Gut. 2009;58(4):501-508. [PubMed: 18832524]

48.

Fidder HH, van de Steen K, van Assche G, Rutgeerts P, Vermeire S. Immortal time bias and infliximab-related mortality and malignancy incidence response. Gut. 2010;59(3):416. [PubMed: 20207649]

49.

Long MD, Kappelman MD, Martin CF, et al. Development of an internet-based cohort of patients with inflammatory bowel diseases (CCFA Partners): methodology and initial results. Inflamm Bowel Dis. 2012;18(11):2099-2106. [PubMed: 22287300]

50.

Randell RL, Long MD, Cook SF, et al. Validation of an internet-based cohort of inflammatory bowel disease (CCFA partners). Inflamm Bowel Dis. 2014;20(3):541-544. [PMC free article: PMC4112538] [PubMed: 24451221]

51.

Lanza ST, Rhoades BL. Latent class analysis: an alternative perspective on subgroup analysis in prevention and treatment. Prev Sci. 2013;14(2):157-168. [PMC free article: PMC3173585] [PubMed: 21318625]

52.

Setoguchi S, Solomon DH, Glynn RJ, Cook EF, Levin R, Schneeweiss S. Agreement of diagnosis and its date for hematologic malignancies and solid tumors between Medicare claims and cancer registry data. Cancer Causes Control. 2007;18(5):561-569. [PubMed: 17447148]

53.

Haynes K, Beukelman T, Curtis JR, et al. Tumor necrosis factor alpha inhibitor therapy and cancer risk in chronic immune-mediated diseases. Arthritis Rheum. 2013;65(1):48-58. [PMC free article: PMC3778442] [PubMed: 23055441]

54.

Kiyota Y, Schneeweiss S, Glynn RJ, Cannuscio CC, Avorn J, Solomon DH. Accuracy of Medicare claims-based diagnosis of acute myocardial infarction: estimating positive predictive value on the basis of review of hospital records. Am Heart J. 2004;148(1):99-104. [PubMed: 15215798]

55.

Kumamaru H, Judd SE, Curtis JR, et al. Validity of claims-based stroke algorithms in contemporary Medicare data: reasons for geographic and racial differences in stroke (REGARDS) study linked with Medicare claims. Circ Cardiovasc Qual Outcomes. 2014;7(4):611-619. [PMC free article: PMC4109622] [PubMed: 24963021]

56.

Hennessy S, Leonard CE, Freeman CP, et al. Validation of diagnostic codes for outpatient-originating sudden cardiac death and ventricular arrhythmia in Medicaid and Medicare claims data. Pharmacoepidemiol Drug Saf. 2010;19(6):555-562. [PMC free article: PMC2924585] [PubMed: 19844945]

57.

Tamariz L, Harkins T, Nair V. A systematic review of validated methods for identifying venous thromboembolism using administrative and claims data. Pharmacoepidemiol Drug Saf. 2012;21(Suppl 1):154-162. [PubMed: 22262602]

58.

Ray WA, Griffin MR, Fought RL, Adams ML. Identification of fractures from computerized Medicare files. J Clin Epidemiol. 1992;45(7):703-714. [PubMed: 1619449]

59.

Grijalva CG, Chung CP, Stein CM, et al. Computerized definitions showed high positive predictive values for identifying hospitalizations for congestive heart failure and selected infections in Medicaid enrollees with rheumatoid arthritis. Pharmacoepidemiol Drug Saf. 2008;17(9):890-895. [PMC free article: PMC4861217] [PubMed: 18543352]

60.

Winthrop KL, Baddley JW, Chen L, et al. Association between the initiation of anti-tumor necrosis factor therapy and the risk of herpes zoster. JAMA. 2013;309(9):887-895. [PMC free article: PMC3773213] [PubMed: 23462785]

61.

Yun H, Xie F, Delzell E, et al. Comparative risk of hospitalized infection associated with biologic agents in rheumatoid arthritis patients enrolled in Medicare. Arthritis Rheumatol. 2016;68(1):56-66. [PubMed: 26315675]

62.

Osterman MT, Haynes K, Delzell E, et al. Effectiveness and safety of immunomodulators with anti-tumor necrosis factor therapy in Crohn's disease. Clin Gastroenterol Hepatol. 2015;13(7):1293-1301.e1295, quiz e1270, e1272. [PMC free article: PMC4475667] [PubMed: 25724699]

63.

Osterman MT, Haynes K, Delzell E, et al. Comparative effectiveness of infliximab and adalimumab for Crohn's disease. Clin Gastroenterol Hepatol. 2014;12(5):811-817 e813. [PMC free article: PMC3883891] [PubMed: 23811254]

64.

Bewtra M, Newcomb CW, Wu Q, et al. Mortality associated with medical therapy versus elective colectomy in ulcerative colitis: a cohort study. Ann Intern Med. 2015;163(4):262-270. [PMC free article: PMC4925099] [PubMed: 26168366]

65.

Gagne JJ, Glynn RJ, Avorn J, Levin R, Schneeweiss S. A combined comorbidity score predicted mortality in elderly patients better than existing scores. J Clin Epidemiol. 2011; 64(7):749-759. [PMC free article: PMC3100405] [PubMed: 21208778]

66.

Hernan MA, Brumback B, Robins JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology. 2000;11(5):561-570. [PubMed: 10955409]

67.

Mamdani M, Sykora K, Li P, et al. Reader's guide to critical appraisal of cohort studies: 2. assessing potential for confounding. BMJ. April 2005;330(7497):960-962. [PMC free article: PMC556348] [PubMed: 15845982]

68.

Attard CL, Brown S, Alloul K, Moore MJ. Cost-effectiveness of folfirinox for first-line treatment of metastatic pancreatic cancer. Curr Oncol. 2014;21(1):e41-e51. [PMC free article: PMC3921047] [PubMed: 24523620]

69.

Romanus D, Kindler HL, Archer L, et al. Does health-related quality of life improve for advanced pancreatic cancer patients who respond to gemcitabine? analysis of a randomized phase III trial of the cancer and leukemia group B (CALGB 80303). J Pain Symptom Manage. 2012;43(2):205-217. [PMC free article: PMC3658140] [PubMed: 22104618]

70.

Ewara EM, Zaric GS, Welch S, Sarma S. Cost-effectiveness of first-line treatments for patients with KRAS wild-type metastatic colorectal cancer. Curr Oncol. 2014;21(4):e541-550. [PMC free article: PMC4117621] [PubMed: 25089105]

71.

Curl P, Vujic I, van 't Veer LJ, Ortiz-Urda S, Kahn JG. Cost-effectiveness of treatment strategies for BRAF-mutated metastatic melanoma. PLoS One. 2014;9(9):e107255. doi:10.1371/journal.pone.0107255 [PMC free article: PMC4157865] [PubMed: 25198196] [CrossRef]

72.

Beusterien KM, Szabo SM, Kotapati S, et al. Societal preference values for advanced melanoma health states in the United Kingdom and Australia. Br J Cancer. 2009;101(3):387-389. [PMC free article: PMC2720221] [PubMed: 19603025]

73.

Delea TE, Amdahl J, Chit A, Amonkar MM. Cost-effectiveness of lapatinib plus letrozole in her2-positive, hormone receptor-positive metastatic breast cancer in Canada. Curr Oncol. 2013;20(5):e371-e387. [PMC free article: PMC3805407] [PubMed: 24155635]

74.

Delea TE, Sofrygin O, Amonkar M. Pcn105 patient preference-based utility weights from the Functional Assessment of Cancer Therapy-General (Fact-G) in women with hormone receptor positive metastatic breast cancer receiving letrozole plus lapatinib or letrozole alone. Value Health. 2010;13(3):A43-A44.

75.

Ramsey SD, Andersen MR, Etzioni R, et al. Quality of life in survivors of colorectal carcinoma. Cancer. 2000;88(6):1294-1303. [PubMed: 10717609]

76.

Cottone M, Kohn A, Daperno M, et al. Advanced age is an independent risk factor for severe infections and mortality in patients given anti-tumor necrosis factor therapy for inflammatory bowel disease. Clin Gastroenterol Hepatol. 2011;9(1):30-35. [PubMed: 20951835]

77.

Lobaton T, Ferrante M, Rutgeerts P, Ballet V, Van Assche G, Vermeire S. Efficacy and safety of anti-TNF therapy in elderly patients with inflammatory bowel disease. Aliment Pharmacol Ther. 2015;42(4):441-451. [PubMed: 26104047]

78.

Peyrin-Biroulet L, Deltenre P, de Suray N, Branche J, Sandborn WJ, Colombel JF. Efficacy and safety of tumor necrosis factor antagonists in Crohn's disease: meta-analysis of placebo-controlled trials. Clinical Gastroenterol Hepatol. 2008;6(6):644-653. [PubMed: 18550004]

79.

Westlake SL, Colebatch AN, Baird J, et al. Tumour necrosis factor antagonists and the risk of cardiovascular disease in patients with rheumatoid arthritis: a systematic literature review. Rheumatology (Oxford). 2011;50(3):518-531. [PubMed: 21071477]

80.

Dixon WG, Watson KD, Lunt M, et al. Reduction in the incidence of myocardial infarction in patients with rheumatoid arthritis who respond to anti-tumor necrosis factor alpha therapy: results from the British Society for Rheumatology Biologics Register. Arthritis Rheum. 2007;56(9):2905-2912. [PMC free article: PMC2435427] [PubMed: 17763428]

81.

van Sijl AM, Boers M, Voskuyl AE, Nurmohamed MT. Confounding by indication probably distorts the relationship between steroid use and cardiovascular disease in rheumatoid arthritis: results from a prospective cohort study. PLoS One. 2014;9(1):e87965. doi:10.1371/journal.pone.0087965 [PMC free article: PMC3907551] [PubMed: 24498229] [CrossRef]

82.

Lewis JD. The utility of biomarkers in the diagnosis and therapy of inflammatory bowel disease. Gastroenterology. 2011;140(6):1817-1826.e1812. [PMC free article: PMC3749298] [PubMed: 21530748]

83.

Mowat C, Cole A, Windsor A, et al. Guidelines for the management of inflammatory bowel disease in adults. Gut. 2011;60(5):571-607. [PubMed: 21464096]

84.

Terdiman JP, Gruss CB, Heidelbaugh JJ, et al. American Gastroenterological Association Institute guideline on the use of thiopurines, methotrexate, and anti-TNF-alpha biologic drugs for the induction and maintenance of remission in inflammatory Crohn's disease. Gastroenterology. 2013;145(6):1459-1463. [PubMed: 24267474]

85.

Baer HJ, Glynn RJ, Hu FB, et al. Risk factors for mortality in the nurses' health study: a competing risks analysis. Am J Epidemiol. 2011;173(3):319-329. [PMC free article: PMC3105270] [PubMed: 21135028]

86.

Gellert C, Schottker B, Brenner H. Smoking and all-cause mortality in older people: systematic review and meta-analysis. Arch Intern Med. 2012;172(11):837-844. [PubMed: 22688992]

87.

Nunes T, Etchevers MJ, Domenech E, et al. Smoking does influence disease behaviour and impacts the need for therapy in Crohn's disease in the biologic era. Aliment Pharmacol Ther. 2013;38(7):752-760. [PubMed: 23980933]

88.

Lawrance IC, Murray K, Batman B, et al. Crohn's disease and smoking: is it ever too late to quit? J Crohns Colitis. 2013;7(12):e665-e671. [PubMed: 23790611]

89.

Askling J, Fahrbach K, Nordstrom B, Ross S, Schmid CH, Symmons D. Cancer risk with tumor necrosis factor alpha (TNF) inhibitors: meta-analysis of randomized controlled trials of adalimumab, etanercept, and infliximab using patient level data. Pharmacoepidemiol Drug Saf. 2011;20(2):119-130. [PubMed: 21254282]

Acknowledgments

The authors acknowledge the contribution of our stakeholders who helped the authors plan the study and interpret the results: Dr James Tarver, PhD; Jackie Spencer, MSW; Dr Robert Baldassano, MD; and Diane Burkhardt.

Research reported in this report was [partially] funded through a Patient-Centered Outcomes Research Institute® (PCORI®) Award (CE-12-11-4143). Further information available at: https://www.pcori.org/research-results/2013/anti-tnf-drugs-versus-long-term-steroid-use-patients-inflammatory-bowel

Appendix

Suggested citation:

Lewis JD, Bewtra M, Scott FI, et al. (2018). Anti-TNF Drugs versus Long-Term Steroid Use for Patients with Inflammatory Bowel Diseases. Patient-Centered Outcomes Research Institute (PCORI). https://doi.org/10.25302/11.2018.CE.12114143

Disclaimer

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.