1 INTRODUCTION

Crohn’s disease (CD) is a chronic, relapsing and progressive form of inflammatory bowel1, 2 disease associated with disabling symptoms3 and often requiring lifelong medical treatments4 and, in many cases, surgery.5 Despite the recent advances in the pharmacological treatment of CD,6 a number of uncertainties still exist regarding the effectiveness and safety of the available biologic therapies and, more importantly, their positioning in a real-life setting. In fact, the available treatment options have rapidly increased over the last two decades since the development of novel monoclonal antibodies. Anti-tumour necrosis factor (TNF) α agents, namely infliximab, adalimumab, and certolizumab, have been used for more than 20 years, and have dramatically improved the management of patients with CD, achieving and prolonging remission,7 presenting an effective option for perianal disease,8 and improving quality of life.9 Vedolizumab, which blocks the α4β7 integrin impeding lymphocyte trafficking (gut homing), was the second class of monoclonal antibodies to be licensed after anti-TNF therapies. Its gut selectivity consequent better safety profile has been demonstrated in long-term extension studies and other data.10 Vedolizumab proved effective and demonstrated a good safety profile in many real-world studies11, 12 with no new safety signals since the registrational trials. The latest monoclonal antibody to be approved for the induction and maintenance of CD was ustekinumab,13 a monoclonal antibody directed against the shared p40 subunit of interleukins (IL)12 and IL23.14 Ustekinumab targets a crucial inflammatory pathway in CD, and is also effective in treating some CD-related extra-intestinal manifestations,15 especially dermatologic and rheumatologic, such as psoriasis and psoriatic arthritis.16 Real-world data exploring its efficacy and safety in CD are still emerging,17-23 as are comparative studies with vedolizumab in anti-TNFα-experienced patients.24-26 In the absence of prospective, randomised clinical trials comparing available treatments, such studies are of particular interest, as they allow comparison after adjustment of potential confounding factors in a real-life setting. The studies published so far have included a limited number of patients and more data are needed.

On this basis, our primary aim was to describe the 3- and 12-month clinical effectiveness of ustekinumab, predictors of clinical response, and its safety profile in a large multicentre cohort of CD patients. As a secondary aim, we performed a propensity score-matched analysis between this cohort of CD patients and a previously described cohort (the Cross Pennine study)11 treated with vedolizumab who failed any anti-TNFα agent.

2 METHODS 2.1 Patients

Eight inflammatory bowel disease (IBD) centres (both university and general hospitals) from the UK took part in this study (Leeds Teaching Hospitals; Manchester Royal Infirmary; Salford Royal Hospitals; Bolton NHS Trust; Bradford Teaching Hospital; The Royal Liverpool and Broadgreen University Hospitals; Wrightington, Wigan and Leigh NHS Trust and The Pennine Acute Hospitals NHS Trust). Data were retrospectively collected from all consecutive adult CD patients (age ≥18), who commenced ustekinumab (January 2017–January 2020; last follow-up available until 30 April 2020) through the electronic medical records. The diagnosis of CD was based according to internationally agreed diagnostic criteria.27, 28 Data of interest were extracted and semi-anonymised from patient records onto a pre-defined spreadsheet that was initially developed for the first Cross Pennine study (MVL),11 and then implemented and centrally collated by the study coordinator (VD) at St. James’s University Hospital, Leeds. Before study initiation, a meeting among all study investigators was held, in order to harmonise data abstraction across multiple sites. A small proportion of the patients recruited from Liverpool have already been described in a previous paper.24 At the time of enrolment, all patients had either gastrointestinal symptoms or objective evidence of inflammation. Patients with incomplete data (i.e., missing medical history, 3- or 12-month clinical assessments not reported, lost to follow-up), uncertain diagnosis or IBD type-unspecified, or with a short follow-up (less than 3 months) were excluded. As per current recommendations, the first dose of ustekinumab was given as an intravenous infusion, diluted in a 0.9% saline solution. The dose was calculated depending on the body weight, according to the approved indication for CD patients. A subcutaneous injection was then administered after 8 weeks, at a dose of 90 mg. Thereafter, patients received a maintenance dose of 90 mg, subcutaneously, at either every 12 or every 8 weeks, as per clinical need, determined by supervising clinicians. Demographic (age, gender) and disease-related data (phenotype according to the Montreal classification, disease duration, previous medications, previous surgery and comorbidities) were collected. Other variables of interest considered were concomitant steroid and immunosuppressive therapy use, bridging steroid therapy, smoking status, previous biologic exposure, and adherence to planned infusions.

For comparison, we have used a cohort of CD patients treated with vedolizumab whose clinical characteristics and clinical outcomes have already been described in the Cross Pennine study,11 after the exclusion of those who had not been exposed to a previous anti-TNF or who had been previously treated with ustekinumab. In brief, these patients have been retrospectively enrolled in the same eight IBD centres between August 2014 and June 2017. Data were retrieved through the local electronic medical records by the treating physicians and all queries were resolved with the study coordinator (MVL). All consecutive patients initiated on vedolizumab in that time span have been included, with the exception of patients who had not yet reached the 3-month clinical assessment. As per current recommendations, vedolizumab was given as an intravenous infusion (300 mg) over 30 minutes at weeks 0, 2, 6, and every 8 weeks thereafter. Also, an additional week 10 dose was administered in 7 patients. The same outcomes were assessed in both the previous11 and the current study, as detailed below.

2.1.1 Outcomes

As for the first Cross Pennine study, the primary outcome of the study was to evaluate the 3- and 12-month short or medium-term clinical effectiveness of ustekinumab and to assess its safety. Clinical response or remission at 14 and 52 weeks (±2 weeks) were assessed using the physician global assessment (PGA) score. PGA outcomes were based on the clinical impression of the treating physician, defining remission as the complete relief or marked improvement of symptoms compared to baseline (score 0), and response as a partial (score 1), though significant, improvement. A PGA score of 2 indicated a moderately active disease, while a score of 3 a severely active disease. In order to support PGA results, the Harvey-Bradshaw Index (HBI) was also included in the statistical analysis, when available, and this was always concordant with the PGA (Spearman’s rho 0.66; P < 0.001). HBI was not used as the primary outcome, as this was not available for all patients. Remission was defined with an HBI score ≤4, while a reduction of at least 3 or more points defined a clinical response. An HBI score 5-7, 8-16, and >16 were indicative of mild, moderate, or severe disease activity. Moreover, data on C reactive protein (CRP) and faecal calprotectin (FC) at baseline and at 14 weeks (±1 week) were collected. CRP was considered as elevated when >5 mg/dL, while FC was considered as significantly increased when >250 µg/g. A number of potential predictors of treatment failure were assessed, including bridging steroid therapy, concomitant use of immunosuppressant (azathioprine, 6-mercaptopurine, or methotrexate), previous anti-TNFα exposure, smoking status, disease duration, baseline FC and CRP, disease phenotype and location, and previous surgery.

Furthermore, we studied treatment discontinuation defined by the treating physician if ustekinumab was judged in their opinion to be ineffective, resulting in cessation of treatment. Specifically, primary failure was defined as inadequate clinical response after induction phase, leading to alternative treatment strategies, also including the need for surgery, while loss of response was defined as inadequate response to treatment occurring any time after the induction phase. Patients who stopped ustekinumab in weeks 14-52 were considered as treatment failure in the 52-week analysis. Data on other reasons for discontinuation, such as adverse events (serious and non-serious), infectious diseases, and other possible safety signals or conditions that were attributed to ustekinumab therapy were also collected. Finally, where applicable, we compared these data with those of the previously described vedolizumab cohort.11

As a secondary aim, we compared the 14- and 52-week remission and response rates according to the PGA in the ustekinumab vs the vedolizumab cohorts using a propensity score-matched analysis. For this purpose, patients who had been previously treated with ustekinumab (in the vedolizumab cohort) or with vedolizumab (in the ustekinumab cohort) were excluded. We have also excluded patients who had not been treated with any anti-TNFα agent as first-line therapy for CD.

As exploratory aims, we also reported 14- and 52-week steroid-free clinical remission, 52-week treatment persistence, 14-week CRP<5 mg/dL, and 52-week hospitalisation before and after matching.

2.2 Statistical analysis

Given the observational, exploratory nature of the study, and in the absence of a pre-defined hypothesis, a sample size was not calculated a priori. A post-hoc calculation of the power showed that this was greater than 80% for the primary outcome, according to the method by Austin based on calculation of the variance inflation factor (VIF), which describes the extent to which the effective sample size has been reduced by weighting.29 The post-hoc power of a chi-square test for the observed difference in proportion of success at 3 months is 0.99, while VIF is 1.15 based on a c-statistics of the model equal to 0.67 and a prevalence of treatment equal to 0.6. Hence, the resulting power is 86%.

Categorical variables were described as count and percentage; quantitative variables as mean and standard deviation (SD) or median and interquartile range (IQR) if not normally distributed. Percentages were calculated after exclusion of patients with missing data, and the analysis of the main outcomes was made on an intention-to-treat basis. Predictors of ustekinumab treatment failure were identified through univariable and multivariable (including only factors with P < 0.05 at univariable analyses) logistic regression models. In order to strengthen our results and to avoid possible biases, these models were also adjusted according to baseline PGA (2-3), inflammatory markers (increased CRP and/or FC), and previous exposure to vedolizumab. Results were expressed as odds ratio (OR) and 95% confidence interval (95% CI). For these subgroup analyses, a P-value below 0.01 was considered significant.

Time to ustekinumab treatment discontinuation (also including comparison with vedolizumab) was represented by the Kaplan-Meier curve for interval data.

The propensity score-matched analysis is widely applied in medical sciences for reducing possible biases, from confounding variables, that may emerge by simply comparing two different treatments.30 In the present study, the aim of the propensity score-matched analysis was to compare the 14- and 52-week clinical remission according to the PGA (score 0-1) between the ustekinumab and the vedolizumab cohorts.

Nearest-neighbour matching (NNM) was applied to the estimated propensity score. The NNM method of treatment-effect estimation imputes the missing potential outcome for each individual by using an average of the outcomes of similar subjects that receive the other treatment level. Each observation is matched with at least 1 observation from the other treatment level. Results are expressed as average treatment effect, computed by taking the average of the difference between the observed and potential outcomes for each subject.

The following pre-treatment variables were selected a priori, based on their clinical relevance as shown in previous studies,24-27 for inclusion in propensity score estimation, namely gender, patient age >65 years, active smoking, patient weight (as a continuous variable), previous CD surgery, colonic only vs small bowel involvement, presence of perianal disease, age at CD diagnosis, presence of extra-intestinal manifestations or psoriasis, disease behaviour, presence of at least one comorbidity other than CD, number of previous anti-TNFα therapies, concomitant immunosuppressive drug (i.e., azathioprine, methotrexate, 6-mercaptopurine), baseline HBI and PGA, baseline CRP and FC, need for steroids at any time as rescue therapy. The MeSH definition was used for defining the presence of comorbidity.31 Psoriasis was included as a relevant variable, given that ustekinumab is the drug of choice in this condition in CD patients who failed anti-TNFα agents.32 We have also performed a sensitivity analysis, only including variables that were significantly associated with the primary outcome at logistic regression.

As the use of steroids was included as a variable of interest, a separate analysis for steroid-free remission was not performed. As a sensitivity analysis, inverse probability weighting (IPW) regression adjustment was also performed.

The overlap assumption that requires that each individual has a positive probability of receiving each treatment was assessed graphically. This graph displays the estimated density of the predicted probabilities that a control patient (vedolizumab) is a control and the estimated density of the predicted probabilities that control is a treated patient (ustekinumab). Balancement instead was assessed through a box plot and by reporting the standardised mean difference (SMD) before and after matching each variable. There was no need for trimming, as the outliers were less than 5%.

The study was performed as a clinical audit using routinely collected clinical data and as such is exempt from the need for ethics committee approval in the UK and the need to take written informed consent. The results of the study are reported according to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) recommendations for quality assurance and according to the Good Research for Comparative Effectiveness (GRACE) initiative.

3 RESULTS 3.1 Cohorts under study

Over the study period, 321 patients with CD commenced treatment with ustekinumab. Of these, we excluded 20 patients who did not reach the 14-week observation, 16 with incomplete or uncertain data, and 3 who were lost to follow-up. Hence, the final ustekinumab cohort included 282 patients (mean age 40 ± 15 years, F:M ratio 1.7:1) who had been treated with this drug for a mean period of 18 ± 5 months, and whose demographic and other relevant characteristics are reported in Table 1, along with the characteristics of the vedolizumab cohort. Almost all patients had already been exposed to a biologic agent, and roughly half of the patients had had CD-related abdominal surgery. Notably, at least one comorbidity was present in 222 patients (78.7%), with essential hypertension, ischaemic heart disease, psoriasis, and type 2 diabetes being the most common.

TABLE 1. Demographic and other relevant characteristics of patients with Crohn’s disease treated with ustekinumab Ustekinumab cohort Vedolizumab cohort N. patients 282 118 Age (years), mean ± SD 40 ± 15 41 ± 16 F:M ratio 1.7:1 1.8:1 Start weight (kg), mean ± SD 71 ± 15 NA Duration of disease (years), mean ± SD 12 ± 9 9 ± 8 Disease phenotype, n (%) A1, 61 (21.6) A1, 34 (32.4) A2, 162 (57.4) A2, 57 (54.3) A3, 59 (20.9) A3, 14 (13.3) L1, 63 (22.3) L1, 14 (13.3) L2, 69 (24.5) L2, 32 (30.5) L3, 150 (53.2) L3, 59 (56.2) B1, 95 (33.7) B1, 40 (38.1) B2, 90 (31.9) B2, 38 (36.2) B3, 97 (34.4) B3, 27 (25.7) Perianal disease, 86 (30.5) Perianal disease, 14 (13.3) Smoking status, n (%) Never smoker 123/238 (51.7) 62/90 (68.9) Former smoker 62/238 (26.1) 10/90 (11.1) Active smoker 53/238 (22.2) 18/90 (20.0) Missing datum, n (%) 44/282 (15.6) 28/118 (23.7) Previous bowel resection, n (%) 131 (46.4) 27 (23.5) At least one comorbidity, n (%) 222 (78.7) 94 (79.7) Extra-intestinal manifestations, n (%) 55 (19.5) 22 (18.6) Psoriasis, n (%) 31 (11.0) 8 (6.8) Any prior biologic drug exposure, n (%) 275 (97.5) 115 (97.4) At least two previous biologics, n (%) 172 (61.0) 70 (59.3) Previous infliximab, n (%) 202 (71.6) 102 (88.7) Previous adalimumab, n (%) 224 (79.4) 79 (67.0) Previous vedolizumab, n (%) 80 (28.4) — Concomitant immunosuppressive therapy, n (%) 93 (32.9) 66 (55.9) Concomitant steroid therapy, n (%) 100 (35.5) 43 (36.4) For the vedolizumab cohort, disease phenotype according to the Montreal classification was not specified in 12/118 (10.2%) cases. Abbreviation: SD, standard deviation. 3.2 Clinical effectiveness

The rates of 14- and 52-week clinical response or remission according to the PGA, as well as steroid-free response or remission, are reported in Figure 1. Clinical response or remission was reached by 200/282 patients (70.9%) at 14 weeks, and by 162/259 patients (62.5%) at 52 weeks. Of note, after induction, in 236 patients (83.7%) ustekinumab was escalated to every 8 weeks. Patients who discontinued ustekinumab and related reasons for discontinuation or loss of response are reported in Table 2. The most common reason for discontinuation was either primary failure or loss of response, followed by the occurrence of adverse events and by the need for surgery. Similar to the previously described vedolizumab cohort, non-adherence rate was rather low (1.4% in the ustekinumab vs 2.4% in the vedolizumab cohort). The Kaplan-Meier ustekinumab failure estimate is reported in Figure S1, while the Kaplan-Meier ustekinumab vs vedolizumab failure estimate is reported in Figure 2. Of note, no significant difference was noticed between the two groups at both 14 and 52 weeks. Finally, Table S1 reports data regarding 14- and 52-week assessment of CRP, FC, and HBI, when available. Of note, HBI and FC significantly decreased at both 14 and 52 weeks.

Rates of 14- and 52-week clinical response or remission and steroid-free response or remission (expressed as %) according to the physician global assessment

TABLE 2. Patients who discontinued ustekinumab and related reasons for discontinuation or loss of response during the study period Discontinuation ≤14 weeks, n (%) 23/282 (8.2) Ineffective (primary non-responder) 15 Adverse effect 6 Need for intestinal surgery 1 Non-adherence 1 Discontinuation >14 and ≤52 weeks, n (%) 49/259 (18.9) Loss of response 32 Adverse effect 6 Need for intestinal surgery 8 Pregnancy 1 Non-adherence 2 Discontinuation >52 weeks, n (%) 29/210 (13.8) Loss of response 15 Achieved remission 2 Need for intestinal surgery 4 Need for non-intestinal surgery 2 Infection 2 Pregnancy 3 Non-adherence 1 Total discontinuations 101/282 (35.8%)

Kaplan-Meier ustekinumab vs vedolizumab failure estimate. Of note, no significant difference was noticed between the two drugs at both 3- and 12-month follow-up

3.3 Predictors of ustekinumab treatment failure

Potential predictors of failure to ustekinumab treatment were assessed in univariable and multivariable analyses, at both 14 and 52 weeks. Besides considering the whole cohort of patients, in order to mitigate the possible bias related to the use of the sole PGA, we also performed alternative analyses pooling together patients with a PGA 2-3 and a PGA 0-1 with either raised CRP or FC (biochemical disease activity). We also performed a separate analysis for patients who had been previously exposed to vedolizumab. All results are reported in Tables S2–S4. Of note, considering the whole sample, at multivariable analysis, high baseline HBI (OR 1.12; 95% CI 1.01-1.24; P = 0.024), Montreal B2 (OR 1.47; 95% CI 0.33-6.47; P = 0.608), and Montreal B3 (OR 3.49; 95% CI 1.01-12.1; P = 0.05) were associated with treatment failure at 14 weeks. At 52 weeks, current smoking, baseline HBI or PGA, and use of steroids were found to be correlated, in almost all sub-analyses (Tables 3 and 4), including vedolizumab-experienced patients, with treatment failure. In the previously published data regarding our vedolizumab cohort, no predictors of treatment failure were found, including bridging steroid therapy, concomitant use of immunosuppressors, smoking status, disease duration, and baseline CRP and FC.11

TABLE 3. Univariable and multivariable analysis of baseline characteristics in relation to the 52-week clinical outcome in the whole sample vs baseline PGA 2-3 in the ustekinumab cohort Whole sample, OR (95% CI), P-value Whole sample, OR (95% CI), P-value Baseline PGA 2-3, OR (95% CI), P-value Baseline PGA 2-3, OR (95% CI), P-value Univariable analysis Multivariable analysis Univariable analysis Multivariable analysis Age 0.99 (0.98-1.01) P = 0.84 — 0.99 (0.97-1.02) P = 0.88 — Female sex 1.34 (0.77-2.33) P = 0.28 — 1.30 (0.62-2.71) P = 0.47 — Disease duration 0.98 (0.96-1.01) P = 0.46 — 0.98 (0.94-1.02) P = 0.37 — Previous smoker 1.53 (0.77-3.03) P = 0.22 — 0.86 (0.34-2.16) P = 0.75 — Current smoker 2.72 (1.36-5.42) P = 0.004 2.48 (1.13-5.44) P = 0.02 2.00 (0.81-4.96) P = 0.13 — Baseline CRP 1.01 (0.99-1.02) P = 0.10 — 1.00 (0.99-1.02) P = 0.27 — Baseline HBI 1.21 (1.12-1.31) P < 0.001 NAa 1.13 (1.02-1.24) P = 0.01 1.13 (1.02-1.25) P = 0.009 Baseline PGA 2.52 (1.72-3.70) P < 0.001 2.4 (1.55-3.69) P < 0.001 — — Increased CRP or FC 0.96 (0.52-1.77) P = 0.90 — 0.84 (0.36-194) P = 0.68 — Montreal L2 0.84 (0.39-1.82) P = 0.67 — 1.14 (0.39-3.27) P = 0.80 — Montreal L3 1.18 (0.61-2.25) P = 0.61 — 0.74 (0.32-1.73) P = 0.49 — Montreal B2 0.64 (0.33-1.25) P = 0.19 — 1.02 (0.43-2.43) P = 0.95 — Montreal B3 1.12 (0.60-2.06) P = 0.71 — 1.31(0.58-2.91) P = 0.50 — Perianal disease 1.53 (0.88-2.68) P = 0.13 — 1.5 (0.72-3.11) P = 0.27 — Previous bowel resection 1.36 (0.80-2.29) P = 0.24 — 1.22 (0.62-2.41) P = 0.55 — Previous infliximab 1.06 (0.60-1.89) P = 0.82 — 0.97 (0.45-2.05) P = 0.93 — Previous adalimumab 2.06 (1.00-4.25) P = 0.04 1.92 (0.82-4.52) p=0.134 1.39 (0.54-3.58) P = 0.48 — Previous vedolizumab 1.24 (0.70-2.19) P = 0.45 — 0.72 (0.34-1.49) P = 0.38 — Concomitant immunosuppressant 0.83 (0.48-1.46) P = 0.53 — 1.05 (0.51-2.15) P = 0.88 — Concomitant steroids 1.90 (1.11-3.25) P = 0.01 2.42 (1.26-4.65) P = 0.008 2.46 (1.21-5.00) P = 0.01 2.27 (1.03-5.00) P = 0.04 Abbreviations: CI, confidence interval; CRP, C-reactive protein; FC, faecal calprotectin; HBI, Harvey-Bradshaw index; NA, not assessable; OR, odds ratio; PGA, physician global assessment. TABLE 4. Univariable and multivariable analysis of baseline characteristics in relation to the 52-week clinical outcome in PGA 2-3 or 0-1 with raised inflammatory markers and in vedolizumab-experienced patients in the ustekinumab cohort Baseline PGA 2-3 or 0-1 with ↑ CRP or FC, OR (95% CI), P-value Baseline PGA 2-3 or 0-1 with ↑ CRP or FC, OR (95% CI), P-value Vedolizumab experienced, OR (95% CI), P-value Vedolizumab experienced, OR (95% CI), P-value Univariable analysis Multivariable analysis Univariable analysis Multivariable analysis Age 1.0 (0.98-1.01) P = 0.87 — 1.01 (0.98-1.04) P = 0.40 — Female sex 1.41 (0.79-2.51) P = 0.24 — 1.66 (0.61-4.47) P = 0.31 — Disease duration 0.99 (0.96-1.02) P = 0.64 — 0.96 (0.90-1.02) P = 0.25 — Previous smoker 1.38 (0.67-2.84) P = 0.37 — 4.66 (1.13-19.1) P = 0.03 3.73 (0.42-33.5) P = 0.24 Current smoker 2.64 (1.28-5.43) P = 0.008 2.32 (1.00-5.37) P = 0.05 5.0 (1.37-18.1) P = 0.01 8.09 (0.78-8.3) P = 0.07 Baseline CRP 1.0 (0.99-1.02) P = 0.16 — 1.0 (0.97-1.03) P = 0.86 — Baseline HBI 1.21 (1.12-1.32) P < 0.001 1.21 (1.10-1.32) P < 0.001 1.28 (1.07-1.53) P = 0.006 1.70 (1.16-2.51) P = 0.007 Baseline PGA — — 1.29 (0.62-2.7) P = 0.48 — Increased CRP or FC — — 1.41 (0.46-4.27) P = 0.53 — Montreal L2 1.0 (0.44-2.25) P = 1.0 — 2.18 (0.34-13.75) P = 0.40 — Montreal L3 1.03 (0.53-2.01) P = 0.91 — 2.57 (0.49-13.5) P = 0.26