Eligible studies were prospective and retrospective cohorts that followed patients for more than 1 year and reported time-to-event outcomes. Events were defined as a discontinuation of anti-TNF therapy for secondary loss of response. Secondary loss of response refers to patients who responded to induction therapy, but subsequently lost response during maintenance treatment. We accepted studies that recorded the Physician's Global Assessment (PGA) of disease activity, as well as studies that used the Pediatric Crohn's Disease Activity Index (PCDAI).

We narrowed our search to studies that exclusively included patients younger than 18 years and were published in English. Papers that were only presented in conferences in the form of an abstract, or those that exclusively focused on patients with perianal or fistulizing CD were excluded. Studies that evaluated the efficacy of anti-TNF agents after bowel resection were also excluded.

We searched for studies published in Medline, Embase, and the Cochrane Library from inception to September 26, 2019. The search strategies for each of the electronic databases are shown in Table 1.

The search results were imported into EndNote X9.2 for de-duplication [5], and subsequently imported in Rayyan, an online screening tool for systematic reviews [6]. One reviewer (HvR) performed the first selection of studies on the basis of title and abstract. The full text manuscript of each potentially eligible study was then obtained. Two authors (HvR and PFvR) independently appraised full text articles against the predefined inclusion criteria. Disagreements were resolved by discussion.

The following characteristics were extracted from each selected study: First author, year of publication, country of origin, cohort definition, observation period, median follow-up time, cohort size, age at study baseline, median time from diagnosis to start of study medication, reason for initiating anti-TNF, induction and maintenance scheme, application of dose escalation and interval shortening, and concomitant medication.

Finally, we extracted time-to-event curve data, as described by Tierney et al. [7] to estimate the probability of continuing biological therapy 1, 2, and 3 years after initiation of anti-TNF therapy. Pooling of time-to-event data was greatly jeopardized due to heterogeneity between studies and was therefore not performed.

This review includes results of electronic searches up to September 26, 2019. In total, 2,464 study titles were screened, of which 94 were selected for full text review (Fig. 1) and 81 were excluded for not meeting the eligibility criteria. A total of 13 papers were ultimately included in the final analysis.

The study characteristics of the included studies are presented in Table 2. Eleven cohort studies evaluated patients who were following IFX therapy [89101112131415161718], while the other two studies described patients who were following ADA therapy [1920]. Two of the eleven studies on IFX had a prospective design [814], while the remainder were retrospective in nature [9101112 1315161718]. All studies were published in the last 10 years, and represent daily practices in North America, Europe, Australia, Israel, and South Korea. In one study, CD patients with perianal involvement were excluded from analysis [12], while in the other twelve studies, perianal involvement varied from 3% [11] to 65% [15]. The cohort size varied between 47 [13] and 502 patients [14], and the median observation period was 1.0 to 3.5 years. The median time from diagnosis to the start of IFX therapy was between 1 month [15] and 1.8 years [9], and from diagnosis to the start of ADA therapy was between 15.7 months [20] and 4.7 years [19]. Patients who received IFX therapy had not been previously exposed to anti-TNFs, while all patients who received ADA therapy had previously failed, or had experienced adverse reactions to, IFX [1920]. In ten cohort studies, a secondary loss of response was defined as a deterioration of PCDAI or PGA [891012141516171920]. In one study, a combination of clinical, biochemical, and endoscopic deterioration was used [13], and in two studies the reason to discontinue anti-TNF therapy was not specified [1118].

We report the first systematic review to incorporate summary time-to-event data to evaluate the long-term efficacy of anti-TNF agents in pediatric CD. To date, there have been two systematic reviews on the short-term efficacy of anti-TNF agents in pediatric patients, one focusing exclusively on IFX [21] and a second exclusively on ADA [22].

Li and colleagues [21] included three RCTs and 19 prospective cohort studies comparing IFX with other therapies (including ADA and exclusive enteral nutrition), and defined induction of endoscopic remission by 14 weeks and maintenance of endoscopic remission at 6 months as their outcomes of interest. Dziechciarz and colleagues [22] were mainly interested in the proportion of patients with remission at 4 and 12 weeks after the first dose of ADA, and the proportion of patients who maintained remission throughout the first year. They included one RCT and 13 observational studies, which were classified as case series; among these studies, we considered two of these to be retrospective cohort studies [1920]. Similar to our study, they also concluded that there is a need for more high-quality evidence on the efficacy of ADA in pediatric CD.

Although the current method provides a means of analyzing time-to-event outcomes for individual studies, it relies solely on published data for systematic reviews. For example, it was not possible to include all relevant studies because the outcome of interest was sometimes missing from the study report. Similarly, this method cannot correct for flaws in the design of the primary studies, such as post hoc exclusion of patients or failure to apply the intention-to-treat principle. In addition, different definitions of secondary loss of response may have affected the comparability of the outcome of interest between studies.

Despite these limitations, real-world evidence studies surely complement the findings from RCTs by providing valuable information on treatment practices and patient characteristics among unselected patients. When individual participant data (IPD) is unavailable, the analysis of time-to-event outcomes is an appropriate alternative.

Trough levels and antibody measurements were not routinely assayed in the studies under review, but rather on a reactive basis when patients had secondary loss of response. In a recently published RCT in pediatric patients with luminal CD, it was shown that proactive measurement of ADA trough levels improved treatment efficacy up to 72 weeks after initiation [23]. These findings are of importance to further optimize the durability of anti-TNF agents in children who are known to be underdosed when standard regimens are applied [24].

The benefits of concomitant immunomodulator use must be balanced against the increased risks of prolonged combination therapy, particularly lymphoma risk with thiopurines. No such risk has been detected when low-dose methotrexate is administered once weekly as an immunomodulator. Stopping concomitant immunomodulation could be considered 6–12 months after initiation provided that IFX trough levels are in range and the patient is in clinical and biological remission.

This review of real-world evidence studies confirms the long-term therapeutic benefit of IFX therapy in children with luminal CD who failed first line therapies. Our findings suggest that combination therapy with an immunomodulator prolongs the durability of IFX therapy. More robust data on ADA therapy are necessary before we can draw definite conclusions about its long-term efficacy.