Second Line Biologic Therapy Following Tumor Necrosis Factor Antagonist Failure: A Real World Propensity Score Weighted Analysis

Abstract

Background and Aims:

Tumor necrosis factor (TNF) antagonists are often used as first line medications to treat moderate to severe inflammatory bowel disease (IBD) but many patients do not achieve or maintain response. Our aim was to compare effectiveness of second line treatments (ustekinumab, vedolizumab, or second TNF antagonist) after TNF antagonist exposure in patients with Crohn’s disease (CD) and ulcerative colitis (UC) from two electronic health records (EHR)-based cohorts.

Methods:

We identified patients with prior TNF antagonist exposure switching to a different biologic in the Mount Sinai Health System (MSHS) EHR (CD: n=527, UC: n=165) and the Study of a Prospective Adult Research Cohort (SPARC) from IBD Plexus (CD: n=412, UC: n=129). Treatment failure was defined as composite of any IBD-related surgery, IBD-related hospitalization, new prescription of oral/intravenous corticosteroids, or need to switch to a third biologic agent. Time-to-event analysis was conducted with inverse probability of treatment weighted data.

Results:

Overall treatment failure occurred in 85% of MSHS and 72% of SPARC CD patients. In SPARC, likelihood of treatment failure was significantly lower with ustekinumab compared to vedolizumab as second line treatment (aHR=0.66, 95%CI 0.54–0.82; p<0.001), a trend confirmed in MSHS (aHR=0.89, 95%CI 0.77–1.04; p=0.15). In both cohorts, superiority of ustekinumab compared to vedolizumab was shown when considering treatment failure as prescription of steroids or third biologic agent. In UC, no differences between second line treatment groups were identified.

Conclusion:

In two independent real-world setting cohorts, second line therapy in CD with ustekinumab after TNF antagonist treatment failure was associated with lower likelihood of treatment failure than second line vedolizumab.

INTRODUCTION

Inflammatory bowel disease (IBD), comprised of ulcerative colitis (UC) and Crohn’s disease (CD), is a relapsing and remitting immune-mediated disease characterized by chronic inflammation and damage of the gastrointestinal tract.^1,2 Patients with moderate to severe IBD are treated with targeted therapies, such as biologics, if conventional therapies do not induce remission.³

Tumor necrosis factor (TNF) antagonists are commonly used first line biologics for the treatment of IBD. However, around 10–30% of patients have primary non-response to TNF antagonist treatment, with an even higher fraction affected by secondary loss of response over time.⁴ In recent years, new drugs with different therapeutic targets, such as α₄β₇ integrin antagonists (vedolizumab) and interleukin-12/23 antagonists (ustekinumab), have been approved.^5–8

Limited head-to-head trials and lack of robust individual prognostic factors for therapeutic response makes the selection of second line medication upon failure of TNF antagonist treatment challenging.⁹ With limited randomized and prospective data on treatment sequencing in IBD, analysis of real-world data can provide initial insights into treatment outcomes. Electronic health records (EHR) allow for the creation of real world patient cohorts with readily available clinical data. In addition, as previously shown, many patients with moderate to severe IBD are not eligible for clinical trials of biologic agents and, therefore, clinical trial populations may not reflect real world patients seen in clinics.¹⁰ We therefore conducted a time-to-event analysis to compare effectiveness of second line therapy following failure of TNF antagonists in patients with IBD using real-world data from two independent EHR-based cohorts.

METHODS

Study populations

We used de-identified structured electronic health record (EHR) data from January 1990 through December 2020 from the Mount Sinai Health System (MSHS) data warehouse, which in total contains data from roughly 9.4 million patients. Clinical history from participants of the Study of a Prospective Adult Research Cohort (SPARC) was extracted from EHR and case report forms.¹¹ We included data until January 2022 from SPARC. Data access to the nationwide SPARC IBD cohort was provided by the Crohn’s and Colitis Foundation via the IBD Plexus program.

Extracted data included diagnosis codes (International Classification of Diseases; ICD-9 and ICD-10), medication prescriptions, procedures (Current Procedure Terminology Fourth Edition; CPT-4 codes, and procedure descriptions), lab test results as well as information on demographics and encounter types of individual patients and events. The study was approved by the Program of Protection for Human Subjects at the Icahn School of Medicine at Mount Sinai.

EHR-based phenotyping of Crohn’s disease and ulcerative colitis

We evaluated different rule-based phenotyping algorithms to define the UC and CD cohort within the MSHS structured EHR data. The approaches included occurrence of at least between one to three coded diagnoses of IBD (ICD-9: 555.XX, 5556.XX; ICD-10: K50.XX, K51.XX) on different dates (“1D”/”2D”/”3D”) in addition to at least one IBD medication prescription (“1M”, Supplementary Table 1).¹² Patients with both CD and UC diagnosis codes in their records were defined as either CD or UC based on consistent diagnosis coding on their five most recent hospital encounters (“5E”). Patients with inconsistent coding were otherwise excluded. We evaluated the phenotyping algorithms based on precision and sensitivity (recall), calculated with clinician verified diagnostic labels from a subset of the MSHS EHR data base, the Mount Sinai Crohn’s and Colitis Registry (MSCCR), which included 893 CD, 516 UC, 23 IBD unclassified, and 368 control cases. For SPARC, clinician verified diagnostic labels for CD, UC, and IBD unclassified were already available and therefore no EHR phenotyping algorithm required.

Identification of biologic therapy sequences

Medication sequencing was determined according to first date of prescription. We excluded patients with less than one year of available clinical data before first biologic prescription, and patients younger than 18 years of age at time of first biologic prescription (Supplementary Figure 1).

Start of first line biologic therapy was considered as first prescription date of any IBD biologic medication (Supplementary Table 1). Second line biologic medication was defined as change in biologic at least 30 days after start of first line TNF antagonist. We grouped together TNF antagonists (infliximab, adalimumab, certolizumab pegol, golimumab; mono-therapy or in combination). Patients with first line TNF antagonist treatment between January 2003 and August 2019 (MSHS) and between July 2004 and May 2021 (SPARC) were included in subsequent analysis.

Outcomes

Our primary outcome was treatment failure defined as a composite of adverse events: any IBD-related surgery, IBD-related hospitalization, IBD-related emergency room visit, new prescription of oral or intravenous corticosteroids, and/or need to switch to a third biologic agent.¹³ We defined any emergency room and inpatient admissions with coded IBD diagnosis (MSHS: exclusion of “secondary”/”related” diagnosis, SPARC: only “primary” diagnosis) as IBD-related hospitalization. For IBD-related surgery, we mapped any SNOMED-CT daughter concept of “Operation on gastrointestinal tract” to CPT-4, ICD-9 and ICD-10-CM codes. For MSHS, IBD diagnosis (not “secondary”/”related” diagnosis) coded for the same visit as the procedure served as indication for IBD-induced surgery, whereas in SPARC, procedure descriptions were based on manual review. Third biologic medication was defined either as change of medication or an added biologic agent to the second line biologic. A gap period between index date (start date of second line medication prescription) and outcome was defined as at least one day for surgery and hospitalization, 30 days for third biologic prescription and 90 days for steroid prescription. Any events occurring before the end of the gap period were not considered. If multiple events of the composite outcome occurred, we considered the time difference between index date and first occurring event date.

We additionally evaluated secondary outcomes of treatment failure defined as any new prescription of oral or intravenous corticosteroids or need to switch to a third biologic agent.

Statistical Analysis

We grouped patients according to first and second line treatments and evaluated baseline variables including sex, age at index date, and race (binarized into non-white and white). Additionally, we included body mass index (BMI), serum albumin, C-reactive protein (CRP), and fecal calprotectin (FC) levels that were reported within 6 months before index date. Further baseline variables of interest were oral and intravenous steroid and immunomodulator prescriptions as well as any IBD-related hospitalization or surgery that occurred between the start dates of first- and second line treatments.

Features with at least 10 counts per treatment group were included for propensity score (PS) estimation. We selected variables based on statistical difference observed at baseline and further included available variables previously known to be associated with therapy response. For the MSHS CD cohort, we considered all above mentioned baseline variables except for FC, for MSHS UC, we additionally removed prior IBD-related surgery. For the PS estimation within the SPARC cohorts, FC and prior IBD-related hospitalization or surgery were excluded due to a limited number of events. We applied median imputation and z-transformation for quantitative variables. We additionally included presence of extra-intestinal manifestations or comorbidities as binary features as sensitivity analysis (Supplementary Table 2–3).¹⁴

For pairwise comparison, PS were estimated using multivariable logistic regression (MatchIt_4.3.2). We calculated inverse probability of treatment weights as previously described.¹⁵ The multinomial PS estimation was based on a tree-based machine-learning model (twang_2.5, mnps function, XGboost model). Diagnostics of PS were based on standardized mean differences of baseline variables between treatment groups before and after weighting the data (twang_2.5, survey_4.1–1).

For time-to-event analyses, we first visualized the data using unweighted Kaplan-Meier plots (survminer_0.4.9) and calculated weighted and unweighted cox proportional hazard models (survival_3.2–13, coxph function). Assumption for proportional hazards was tested and no violation was identified. Inverse probability of treatment weighting (IPTW) was applied as weights. Additionally, IPTW-weighted parametric survival regression based on the Weibull distribution was modeled (survival_3.2–13,). We tested whether the data follow Weibull distribution and reported on results if violation of the assumption was not identified (fitdistrplus_1.1–6, and Kolmogorov-Smirnov Tests).

RESULTS

Definition of IBD cohorts at Mount Sinai

We defined our cohorts using at least three IBD diagnoses, one IBD medication, and consistent coding on up to the five most recent encounters. These criteria yielded high precision (CD: 0.93, UC: 0.94), sensitivity (recall; CD: 0.87, UC: 0.76), specificity (CD: 0.97, UC: 0.98), and patient counts (CD: n=10,030, UC: n=6,911) when defining the MSHS CD and UC cohorts (Supplementary Table 4, Figure 1). After exclusion of patients with unknown sex and unknown date of birth, the MSHS cohorts consisted of 9,804 CD and 6,696 UC patients (Supplementary Table 5).

Figure 1:

Crohn’s disease (CD) and ulcerative colitis (UC) cases were defined based on diagnosis, medication, and encounter information. Patients with at least one IBD medication prescription, three IBD diagnosis codes on different dates, and consistent CD or UC diagnosis codes on the most recent five encounters were classified as CD or UC cases, respectively (green). If no criteria was fulfilled, patients were classified as controls (yellow), otherwise excluded (red).

Biologic agents were prescribed to 57% (n=5,544) of CD and 27% (n=2,162) of UC patients within the MSHS cohort. Most commonly prescribed biologics in CD were infliximab (n=2,857), adalimumab (n=2,511), and ustekinumab (n=1,144) in UC infliximab (n=1,257), vedolizumab (n=816), and adalimumab (n=641).

Treatment success of second line biologic medication in CD

We identified 527 and 412 CD patients prescribed second line biologics following TNF antagonist exposure and met follow-up time requirements from MSHS and SPARC, respectively. Median follow-up time was 3.2 years in MSHS and 3.8 years in SPARC. The most commonly prescribed first line TNF antagonist was adalimumab (MSHS n=238, SPARC n=208), followed by infliximab (MSHS n=230, SPARC n=147) and certolizumab pegol (MSHS n=53, SPARC n=46) (Supplementary Table 6). Overall, failure of second line medication was observed in 85% of the MSHS CD cohort and 72% of the SPARC CD cohort (Supplementary Figure 2–3, Supplementary Table 7–8).

We compared effectiveness of second line treatment with ustekinumab (MSHS n=145, SPARC n=140), vedolizumab (MSHS n=74, SPARC n=73) and second TNF antagonist agent (MSHS n=303, SPARC n=150) following first line treatment with TNF antagonists. Unweighted Kaplan-Meier plots indicated superior outcomes for CD patients treated with second line ustekinumab compared to vedolizumab or second TNF antagonist in both the MSHS as well as SPARC cohorts (Figure 2A,B,D,E).

Figure 2:

Unweighted Kaplan-Meier analysis comparing success of second line treatment following failure of TNF antagonist treatment with vedolizumab, ustekinumab and TNF antagonists in (A) MSHS and (B) SPARC CD cases, and with ustekinumab and TNF antagonists in (C) MSHS and (D) SPARC UC cases. P-Value of a log-rank test is plotted.

After adjusting data with IPTW propensity scoring weights, the standardized difference in baseline variables between the different treatment groups was less or equal than 0.1 for all variables in both CD cohorts (Table 1). On adjusted analysis in the SPARC cohort, both second line ustekinumab (adjusted hazard ratio [aHR]=0.66, 95% CI 0.54, 0.82; p<0.001) and TNF antagonist (aHR=0.72, 95% CI 0.60, 0.87; p<0.001) treatment showed significantly lower risk of failure compared to vedolizumab. In the MSHS cohort, second line ustekinumab, compared to vedolizumab, was not significantly associated with a difference in treatment failure risk, but followed same directionality as SPARC (HR=0.89, 95% CI 0.77–1.04; p=0.15). No difference in treatment success was identified between second line vedolizumab and TNF antagonist therapy (aHR=0.99, 95% CI 0.86–1.14; p=0.89) (Table 2).

Table 1.

Baseline Variables in the MSHS and SPARC CD Second-Line Biologic Treatment Groups After Failure of TNF Antagonist Treatment

	MSHS						SPARC
	Vedolizumab	Ustekinumab	TNF antagonist	P	SMD	SMD IPTW	Vedolizumab	Ustekinumab	TNF antagonist	P	SMD	SMD IPTW
n	74	145	303				72	139	150
Sex, male (%)	33 (44.6)	64 (44.1)	142 (46.9)	.84	0.04	0.01	38 (52.8)	54 (38.8)	52 (34.7)	.03	0.25	0.06
Age at index, y, median [IQR]	42.00 [31.25–53.75]	43.00 [32.00–55.00]	37.00 [29.00–48.50]	.00	0.21	0.06	40.00 [29.75–56.00]	41.00 [29.00–53.00]	38.00 [30.00–48.00]	.64	0.08	0.01
Race, White (%)	59 (79.7)	117 (80.7)	229 (75.6)	.43	0.08	0.05	62 (86.1)	119 (85.6)	124 (82.7)	.72	0.06	0.04
Serum albumin, median [IQR]a	3.80 [3.46–3.96]	3.80 [3.63–3.90]	3.80 [3.60–4.00]	.39	0.12	0.03	4.00 [4.00–4.12]	4.00 [4.00–4.00]	4.00 [3.80–4.00]	.07	0.17	0.10
BMI, median [IQR]a	−0.28 [−0.78 to 0.09]	−0.18 [−0.59 to0.57]	−0.18 [−0.59 to0.45]	.07	0.22	0.08	−0.14 [−0.55 to0.40]	−0.14 [−0.40 to0.40]	−0.14 [−0.65 to0.11]	.38	0.13	0.10
CRP, median [IQR]a	6.95 [4.41–11.85]	6.95 [3.85–11.70]	6.95 [5.23–11.35]	.51	0.12	0.10	1.80 [1.80–1.95]	1.80 [1.80–1.80]	1.80 [1.80–1.80]	.71	0.08	0.06
Hospitalization (%)b	43 (58.1)	96 (66.2)	214 (70.6)	.11	0.18	0.09	0 (0.0)	4 (2.9)	7 (4.7)	.40	0.13	–
Immunomodulator prescription (%)b	30 (40.5)	63 (43.4)	150 (49.5)	.26	0.12	0.08	30 (41.7)	67 (48.2)	77 (51.3)	.40	0.13	0.06
Steroid prescription (%)b	44 (59.5)	78 (53.8)	163 (53.8)	.66	0.08	0.04	48 (66.7)	82 (59.0)	85 (56.7)	.36	0.14	0.05
Surgery (%)b	22 (29.7)	38 (26.2)	88 (29.0)	.79	0.05	0.06	0 (0.0)	4 (2.9)	4 (2.7)	.36	0.16	–
EIMs (%)	10 (13.5)	25 (17.2)	61 (20.1)	.384	0.12	–	10 (13.9)	18 (12.9)	19 (12.7)	.968	0.02	–
Comorbidities (%)	43 (58.1)	77 (53.1)	172 (56.8)	.705	0.067	–	37 (51.4)	80 (57.6)	58 (38.7)	.005	0.256	–

NOTE. For categoric variables, differences were tested using the chi-squared test. For numeric values, comparisons were conducted using the Kruskal–Wallis rank-sum test. Standardized mean differences (SMDs) are given before and after inverse probability of treatment weighting (IPTW).

CD, Crohn’s disease; EIM, extraintestinal manifestation; IQR, interquartile range; MSHS, Mount Sinai Health System; SPARC, Study of a Prospective Adult Research Cohort; TNF, tumor necrosis factor.

^aThe latest values of serum albumin, C-reactive protein (CRP), and body mass index (BMI) were considered, no longer than 6 months before the first prescription of second-line therapy.

^bMedication prescription, IBD-related hospitalization, and surgery were considered between the first prescription of first-line therapy and the first prescription of second-line therapy.

Table 2:

Adjusted Hazard ratio (aHR) of treatment success using ustekinumab or TNF antagonist therapy compared to vedolizumab as second line medication upon failure of TNF antagonist treatment in the MSHS and SPARC CD cohorts.

Cohort	Second line medication	aHR [95% CI]	aHR p-value
MSHS	Ustekinumab (ref: vedolizumab)	0.89 [0.77, 1.04]	0.15
SPARC	Ustekinumab (ref: vedolizumab)	0.66 [0.54, 0.82]	<0.001
MSHS	TNF antagonist (ref: vedolizumab)	0.99 [0.86, 1.14]	0.89
SPARC	TNF antagonist (ref: vedolizumab)	0.72 [0.60, 0.87]	<0.001

Time-to-event analysis is based on IPTW-adjusted regression models using Weibull distribution.

We additionally conducted time-to-event analysis for steroid and third biologic prescription as individual outcomes of the composite (Supplementary Table 9). The hazards for steroid and third biologic prescription were significantly lower for ustekinumab in comparison to vedolizumab in the SPARC cohort (new steroid: aHR=0.64, 95% CI 0.42–0.96, p=0.03; third biologic: aHR=0.43, 95% CI 0.32–0.57, p<0.001) with similar results in MSHS (new steroid: aHR=0.57, 95% CI 0.35–0.92, p=0.02; third biologic: aHR=0.60, 95% CI 0.48–0.79, p<0.001).

Treatment success of second line biologic medication in UC

165 and 129 UC patients with prior TNF antagonist exposure from MSHS and SPARC, respectively, were prescribed second line biologics and had a minimum follow-up time of six months after start of second line biologic treatment. Median follow-up time was 2.9 years in MSHS and 3.4 years in SPARC. In MSHS, infliximab was most commonly prescribed as first line TNF antagonist (n=98), followed by adalimumab (n=58), whereas in SPARC, adalimumab was more commonly prescribed (n=66) compared to infliximab (n=59; Supplementary Table 6). Based on the composite outcome definition, 79% of MSHS UC and 69% of SPARC UC patients had treatment failure (Supplementary Figure 4–5, Supplementary Table 7–8). We compared second line treatment with vedolizumab (MSHS n=72, SPARC n=54) and second TNF antagonist (MSHS n=79, SPARC n=44) after first line TNF antagonist treatment. Unweighted Kaplan-Meier plots did not demonstrate any differences in treatment failure rates when comparing second line biologic treatments in both UC cohorts (Figure 2E,F).

After IPTW propensity scoring weight adjustment the groups were well balanced with minor differences in baseline albumin levels (SMD=0.14) and CRP (SMD=0.16) in MSHS UC (Table 3). We did not detect differences in treatment failure with second line vedolizumab compared to second line TNF antagonist treatment in adjusted time-to-event analyses in both MSHS (aHR=1.03, 95% CI 0.62–1.71; p=0.91) and SPARC (aHR=0.86, 95% CI 0.43–1.72; p=0.66) cohorts (Table 4). None of the individual nor composite outcome results in the UC cohorts indicated significant differences between the two treatment groups (Supplementary Table 9).

Table 3:

Baseline variables in the MSHS and SPARC UC and second line treatment groups after failure of TNF antagonist treatment. For categorical variables, differences were tested using Chi-squared test. For numeric values, comparisons was conducted using the Man-Whitney U test. Standardized mean differences (SMD) are given before and after inverse probability of treatment weighting (IPTW).

	MSHS					SPARC
	Vedolizumab	TNF antagonist	p	SMD	SMD IPTW	Vedolizumab	TNF antagonist	p	SMD	SMD IPTW
n	72	77				54	44
Sex (male, %)	33 (45.8)	36 (46.8)	1.00	0.02	0.06	19 (35.2)	24 (54.5)	0.09	0.40	0.00
Age at index (median [IQR])	39.50 [30.75, 58.00]	43.00 [32.00, 55.00]	0.69	0.05	0.04	42.00 [31.00, 54.75]	41.50 [29.00, 54.25]	0.68	0.10	0.03
Race (white, %)	55 (76.4)	47 (61.0)	0.07	0.34	0.01	46 (85.2)	40 (90.9)	0.58	0.18	0.02
Serum albumin (median [IQR]) *	3.88 [3.55, 4.04]	3.88 [3.48, 4.20]	0.55	0.09	0.14	4.00 [4.00, 4.10]	4.00 [3.80, 4.03]	0.29	0.30	0.02
BMI (median [IQR]) *	−0.23 [−0.82, 0.51]	−0.23 [−0.57, 0.63]	0.29	0.18	0.02	−0.10 [−0.57, 0.26]	−0.10 [−0.79, 0.44]	0.82	0.06	0.01
CRP (median [IQR]) *	4.44 [1.96, 5.74]	4.44 [2.30, 12.25]	0.16	0.39	0.16	1.05 [0.94, 1.68]	1.05 [0.45, 1.38]	0.59	0.04	0.01
Hospitalization (%) **	38 (52.8)	46 (59.7)	0.49	0.14	0.00	1 (1.9)	0 (0.0)	1.00	0.19	-
Immunomodulator prescription (%) **	33 (45.8)	34 (44.2)	0.97	0.03	0.03	20 (37.0)	13 (29.5)	0.57	0.16	0.03
Steroid prescription (%) **	47 (65.3)	45 (58.4)	0.49	0.14	0.04	34 (63.0)	23 (52.3)	0.39	0.22	0.01
Surgery (%) **	6 (8.3)	11 (14.3)	0.38	0.19	-	2 (3.7)	0 (0.0)	0.57	0.28	-
EIMs (%)	8 (11.1)	13 (16.9)	0.438	0.27	-	8 (14.8)	5 (11.4)	0.84	0.10	-
Comorbidities (%)	32 (44.4)	43 (55.8)	0.22	0..23	-	30 (55.6)	20 (45.5)	0.428	0.20	-

^*Latest values of serum albumin, C-reactive protein (CRP) and body-mass-index (BMI) considered, no longer than 6 months before first prescription of second line therapy.

^**Medication prescription, IBD-related hospitalization and surgery considered between first prescription of first line therapy and first prescription of second line therapy.

Table 4:

Adjusted Hazard ratio (aHR) of treatment success using vedolizumab instead of TNF antagonist therapy as second line medication upon failure of TNF antagonist treatment in the MSHS and SPARC UC cohorts.

Cohort	Second line medication	aHR [95% CI]	aHR p-value
MSHS	TNF antagonist (ref: vedolizumab)	1.03 [0.62, 1.71]	0.91
SPARC	TNF antagonist (ref: vedolizumab)	0.86 [0.43, 1.72]	0.66

Time-to-event analysis is based on IPTW-adjusted regression models using Weibull distribution.

Last, in both CD and UC cohorts, we conducted a number of sensitivity analysis removing patients with off-label drug prescriptions before FDA approval or including presence of extra-intestinal manifestations and comorbidities into the PS estimation, the results remained unchanged (Supplementary Table 10).

DISCUSSION

In this study using two independent real-world, EHR-based cohorts, we observed that patients with CD with prior TNF antagonist failure had better clinical outcomes with second line ustekinumab compared to second line vedolizumab or alternate TNF antagonist. No difference in effectiveness was observed when comparing second line vedolizumab with starting an alternate TNF antagonist in UC patients with prior TNF antagonist failure.

Our data add to prior limited observational and network meta-analysis data on optimal biologic sequencing in IBD. Here, we replicated our analysis based on the CD and UC cohorts from the MSHS in the SPARC cohorts, yielding large cohorts compared to previously published studies. In previous studies based on retrospective cohorts from IBD centers in the UK, the Netherlands, and France, it was observed that ustekinumab is likely more effective as second line medication compared to vedolizumab in TNF antagonist failure CD patients.^16–19 The information on whether increased efficacy relates to the induction or maintenance phase is conflicting. A recent network meta-analysis by Singh et al. indicates that the odds to induce remission in CD patients with previous biologic exposure is higher for ustekinumab, adalimumab, and riskankizumab compared to vedolizumab.²⁰ Parrot et al. suggested in their meta-analysis that ustekinumab and vedolizumab are similarly effective in inducing remission in CD patients who did not have response to TNF antagonist medication, whereas ustekinumab is more effective to maintain clinical remission.²¹ Importantly, the outcomes of prior studies were often based on clinical or endoscopic remission at specific time points. Therefore, it is difficult to directly compare our study with prior ones, but the general observations are consistent in favoring ustekinumab over vedolizumab as second line therapy. As previously described, TNF antagonist resistance may in part be driven by IL-23 pathways and cells, potentially making ustekinumab a suitable choice of medication in TNF-refractory patients.²² The remission rates with vedolizumab in TNF exposed CD patients are low according to clinical trial data.²³

Fewer data are available regarding effectiveness of second line biologics in UC. While one study based from the Swedish health registry suggested that effectiveness and safety of second line TNF antagonist compared to vedolizumab after TNF antagonist failure is similar²⁴, another study based on a retrospective observational cohort from eight Italian IBD referral centers identified higher risk of therapeutic failure at week 52 for adalimumab compared to vedolizumab after failed infliximab treatment.²⁵ In a network meta-analysis, ustekinumab and tofacitinib were found to be more effective at inducing clinical remission compared to vedolizumab and adalimumab in UC patients with prior TNF antagonist exposure.²⁶ According to Sands et al., vedolizumab may best be positioned in UC patients without prior TNF antagonist exposure.²⁷

Our study’s strengths include use of two independent real-world cohorts and IPTW-adjusted data for baseline likelihood of medication prescriptions. In addition, we demonstrate the feasibility of interrogating large EHR based cohorts to investigate biologic sequencing in IBD. We also acknowledge a number of limitations. First, as any observational study from real-world data, there is risk of confounding by indication. The automated phenotyping algorithm we applied to identify UC and CD cases from the MSHS EHR data yielded very high accuracy but could potentially be improved further by including information from clinical notes or medical chart review. Even though we adjusted the data for known potential confounders, we cannot exclude the possibility of confounding effects from unknown or incompletely known variables.²⁸ With the IPTW method we only considered known factors that are extractable from structured EHR. Furthermore, information on reason for termination of TNF antagonist treatment was not reliably recorded in the EHR which may impact effectiveness of second line TNF antagonist treatment (i.e. second line TNF antagonist may be more effective if reason for first line treatment failure was immunogenicity).²⁹ The exact reason for termination of second line treatments (e.g. due to adverse events) could not be reliably ascertained. Additionally, we were unable to include certain granular clinical information that is not coded in the EHR such as clinical-based disease activity scores, endoscopic scores or histology into our models. The SPARC EHR cohort primarily includes outpatient data. Hence, there is a likelihood of missing IBD-related hospitalization and IBD-related surgeries, likely explaining the lower fraction of treatment failure in the SPARC cohorts compared to MSHS. We used additional clinical data sources from SPARC provided as case report forms upon patient enrollment to supplement EHR data. We mitigated incompleteness of the data by using a composite outcome definition of treatment failure including medication prescriptions. For the MSHS and SPARC cohorts, we excluded patients with less than one year of data before their first targeted therapy prescription, to reduce likelihood of falsely identifying first and second line biologics. Further, we excluded patients with less than six months follow-up time to increase chances of capturing failure of treatment. Nevertheless, we potentially missed events from the composite outcome definition that might have occurred at a later time point. Last, there were insufficient data on newer small molecule therapies, such as tofacitinib, to incorporate these potential second line therapies in our study.

In conclusion, using two large EHR-based cohorts, our data suggests that ustekinumab may be preferred second line therapy compared to vedolizumab in CD patients with prior TNF antagonist failure. . In contrast, there was no clear benefit of using vedolizumab or an alternate TNF antagonist in patients with UC with prior TNF antagonist failure.

What You Need To Know:

BACKGROUND:

TNF antagonists are commonly prescribed in moderate to severe inflammatory bowel disease, but loss of response is common. Here, we compared treatment success of second line biologic treatment upon failure of TNF antagonist therapy.

FINDINGS:

Second line therapy with ustekinumab, compared to vedolizumab, was less likely to fail in patients with Crohn’s disease with prior TNF exposure in two independent electronic health record-based cohorts.

IMPLICATIONS FOR PATIENT CARE:

These data suggest that ustekinumab should be preferred over vedolizumab as a second line treatment in Crohn’s disease upon failure of TNF antagonists.

Abbreviations:

aHR

adjusted Hazard Ratio

BMI

Body Mass Index

CI

Confidence Interval

CPT-4

Current Procedure Terminology Fourth Edition

CRP

C-Reactive Protein

EHR

Electronic Health Records

FC

Fecal Calprotectin

ICD

International Classification of Diseases

IPTW

Inverse Probability of Treatment Weighting

MSCCR

Mount Sinai Crohn’s and Colitis Registry

MSHS

Mount Sinai Health System

PS

Propensity Score

SMD

Standardized Mean Difference

SPARC

Study of a Prospective Adult Research Cohort

TNF

Tumor Necrosis Factor

Data Transparency Statement:

Data and supporting materials from MSHS cohort will be made available to other researchers upon reasonable request and the approval from the corresponding authors. The clinical data from the SPARC study was provided by the IBD Plexus Program of the Crohn’s & Colitis Foundation and can be accessed upon application and approval.