Systematic review with meta‐analysis: effectiveness of anti‐inflammatory therapy in immune checkpoint inhibitor‐induced enterocolitis

Immune checkpoint inhibitors have revolutionised cancer treatment, but at the cost of off‐target immune‐mediated organ damage. This includes checkpoint inhibitor‐induced enterocolitis which frequently requires hospitalisation and may be life‐threatening. Empiric treatment typically includes corticosteroids and infliximab, although no large‐scale studies have confirmed their effectiveness.

Collins et al 17 published a recent systematic review which offered insights into the management of this evolving mucosal disease.
The current study complements this work, by providing an updated systematic review and the first meta-analysis with meta-regression to quantify the efficacy of anti-inflammatory therapy in checkpoint inhibitor-induced enterocolitis. Potentially relevant papers were obtained and evaluated in detail, with the reference lists used to carry out a recursive search of the literature. Articles were assessed independently by two investigators (HI and MAS) according to the predefined eligibility criteria.

| Search strategy
Any disagreement between investigators was resolved by consensus or discussion with a third investigator (NP), if a consensus was not reached.

| Outcomes of interest
In this study, the term 'checkpoint inhibitor-induced enterocolitis' is used to denote inflammation of the gastrointestinal tract that is usually associated with diarrhoea.
The inclusion criteria included: adult patients with any solid or haematological malignancy receiving at least one dose of any checkpoint inhibitor; availability of data for rate of checkpoint inhibitor-induced enterocolitis and response to anti-inflammatory therapy and studies where more than 5 patients received anti-inflammatory therapy. Studies where checkpoint inhibitor therapy was delivered in combination with other therapies (eg radiotherapy, chemotherapy, etc.) were excluded.
For studies that were close to fulfilling the inclusion criteria, corresponding authors were contacted by email to see if additional data were available that might qualify the study as eligible. For example, in studies where there was ambiguity in the number of patients responding to anti-inflammatory therapy.
The main outcomes of interest included the anti-inflammatory agent used (with regimen and dose-if available) and number or proportion of patients 'responding' to anti-inflammatory therapy.
'Response' was taken to be the definition used by authors in their respective studies.

| Data extraction
Data were extracted by two authors in duplicate, according to a predefined protocol and recorded in a Missing outcome data for patients who had received anti-inflammatory therapy were excluded from analysis.

| Assessment of study quality
The methodological quality of studies was evaluated using a quality appraisal tool by Moga et al. 18 This uses an 18-point checklist to evaluate the domains of study objective, study population, interventions and co-interventions, outcome measures, statistical analysis, results and conclusions and competing interests. Studies were awarded points according to preset criteria agreed between two authors (Appendix B).

| Data synthesis and statistical methods
Due to substantial between-study variance in effect size, a random effects meta-analysis model, according to the method of DerSimmonian and Laird, 19 was used to calculate the pooled estimate proportion of checkpoint inhibitor-induced enterocolitis patients who responded to each anti-inflammatory agent (rounded to the nearest significant figure). Each proportion was logit transformed prior to analysis and then the pooled estimate and 95% confidence interval boundaries were back transformed to a proportion scale using the antilogit formula. Heterogeneity was assessed using the I 2 method, with a threshold of ≥50% to define a substantial heterogeneity, 20 and the Cochrane chi-squared test with a P ≤ 0.10, used to define a significant degree of heterogeneity. 26 Comparisons between checkpoint inhibitor regimens and between cancer types were performed using random-effects meta-regression. 21  Some studies included cohorts with a mixed population of patients on a range of checkpoint inhibitor regimens and across a range of cancers, with limited individual data available. For the subgroup analysis, it was agreed that when at least 75% of the study population had a particular cancer or were on a particular checkpoint inhibitor regimen, they were included in that respective subgroup category for analysis (eg if one study had 80% of patients treated with an anti-CTLA-4 containing regimen, that study was included in the anti-CTLA-4 group). If a study did not fulfil this criterion, it was depicted on the forest plot under the label 'mixed'.
Funnel plots were produced for the principal outcome for each comparison, and Egger's test of funnel plot asymmetry was used to assess publication bias. 27 All statistical analyses described above were performed using the STATA (version 16) software. Reporting of this study adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement.

| Search
After de-duplication, the search strategy identified a total of 4093 citations, of which 98 were potentially relevant and retrieved for further assessment (Figure 1). Of these, 59 were excluded for various reasons leaving 39 eligible studies (Table 1). Thirteen were abstracts and 26 were full articles. One article was translated from Japanese. 22 The majority of studies were either observational studies (prospective and retrospective) or case series. Detailed characteristics of included studies are shown in Table 1. Of note there were no discrepancies between authors, in terms of studies deemed suitable for inclusion.
Corticosteroids and infliximab were the most frequently administered anti-inflammatory agents.
All but one study 22

| Study quality
Studies scored between 5/18 and 17/18 points on the quality assessment for case series checklist (see Appendix B), with an average score of 11. Studies scoring lower marks were predominantly abstracts which lacked the detail to achieve points in the relevant categories.
Points were also frequently deducted for studies being single centre, and lack of reporting for length of follow up and adverse events.

| Publication bias
Funnel plots did not show significant asymmetry for any cohorts (cancer type, checkpoint inhibitor regimen) for either corticosteroids or infliximab, with Eggers test P value ≥ 0.5 in all cases. Overall, the pooled response to corticosteroids was 59% (95% CI 54-65), which was associated with a high degree of statistically significant heterogeneity between studies (I 2 = 61%, P < 0.001) ( Figure 2A). The median time to response was defined in seven stud-  Not included in meta-analysis as n

| Efficacy of infliximab
Seventeen studies reported outcomes of infliximab therapy in 333 patients not achieving an adequate response to corticosteroids. The exception was in the O'Connor et al study 35 where one patient was given infliximab as primary therapy due to "severity of symptoms" and a previous serious adverse effect related to corticosteroid use.
Over half of the pooled cohort was contributed by 5 studies, Foppen In most studies, the number of infusions varied between one and three depending on the clinical response, although two studies were less specific using either 'more than one dose' 10 or 'more than two doses'. 36 Figure 3B). The remainder studies included 'mixed' or mainly 'unknown' cancer cohorts. The pooled response rate in melanoma-treated patients was 74% (95% CI 64-82), which was associated with a moderate degree of statistically significant heterogeneity (I 2 37%, P = 0.1), and thus may partly account for the overall heterogeneity seen.

| Efficacy of vedolizumab
Three studies (two articles and one abstract) reported outcomes of vedolizumab in checkpoint inhibitor-induced enterocolitis, 43 In the larger Abu-Sbeih et al study, 43  There were too few studies to perform subgroup analysis or meta-regression. Interestingly, the efficacy of corticosteroids in checkpoint inhibitor-induced enterocolitis is broadly comparable to response rates observed in patients with acute, severe ulcerative colitis. [46][47][48] A potentially important and previously unrecognised insight from our study, was that clinical response to corticosteroids was signifi- A tumour-specific effect on incidence of checkpoint inhibitorinduced enterocolitis has been described, with melanoma patients appearing to have higher rates of colitis. 54 This prompted us to probe whether a differential response to anti-inflammatory therapy may also be linked to tumour type. The current study shows that cancer type-namely, melanoma and lung cancer-were significant moderators in effect size, with melanoma patients experiencing a less favourable response to corticosteroid therapy.

| D ISCUSS I ON
Another important finding pertains to the clinical utility of budesonide-a topical corticosteroid that has a well documented role in the management of IBD. Although prophylactic budesonide, was not effective in preventing ipilimumab-induced diarrhoea in a phase 2 randomised placebo-controlled trial, 49 our study highlights the potential for budesonide as a primary therapeutic strategy in checkpoint inhibitor induced-enterocolitis. Eight studies report the use of budesonide, 10,14,16,27,29,30,33,37 although few describe the clinical outcomes related to its administration. De Felice et al report that 5 of 6 prednisolone refractory patients achieved a 'complete response' after treatment with budesonide (9-12 mg). 30 Hughes et al used budesonide to successfully treat over 50% (exact rate not extractable) of 12 patients with checkpoint inhibitor-induced microscopic colitis. 33 Other studies report a few cases where budesonide was successfully used as the primary corticosteroid. 27,29 In line with this, a small case series (n = 2) also demonstrated that Infliximab has an established role as second-line therapy in corticosteroid refractory cases, or some instances of corticosteroid relapse. 17,51 Other anti-TNF agents have not been extensively studied, although successful use of adalimumab has been reported. 15,52 There is evidence that timely initiation of infliximab is associated with a shorter time to resolution, shorter duration on corticosteroids as well as lower rates of recurrence. 9 In terms of dose, 5 mg/kg has been widely adopted, with the number of infusions administered seeming to depend on clinical response, but generally not exceeding 3 doses, before alternative therapeutic strategies are sought.
There is a movement towards adopting a 'top-down' approach, as used in the IBD paradigm. One study introduced early infliximab to checkpoint inhibitor-induced enterocolitis patients in a predefined schedule, administering three or more infusions regardless of response to corticosteroids. 43 This was associated with a reduced length of hospital stay, reduced need for re-hospitalisation, increased likelihood of successful corticosteroid taper and a lower recurrence rate compared to patients who received less than three infliximab infusions, and at a later time course in their disease.
A differential response to infliximab based on checkpoint inhibitor regimen was less apparent than in the corticosteroid subgroup, although this was challenging to ascertain given there was only one anti-PD-1/L1 treated study included. 42 Notably, all 6 infliximab treated patients experienced a response, 42 compared pneumonia. 25 An infusion reaction after the second dose of infliximab was reported. 37 Three studies report an absence of 'serious' or 'major' adverse effects. 10,38,40 It is worth highlighting the challenges of extrapolating the infection risk incurred by infliximab in this context, given the additive effect of corticosteroids.
In relation to vedolizumab, one study noted an absence of adverse events. 44 The 2018 vedolizumab Abu-Sbeih et al study, which included patients from the cohort in their most recent series, describes one patient who developed a skin rash and another with diffuse joint pain after one dose leading to discontinuation. 53 While our study focused on the most widely used agents currently used to treat checkpoint inhibitor-induced colitis, other therapies such as 5-aminosalicylic acid (5-ASA), calcineurin inhibitors and mycophenolate mofetil (MMF) have been described, albeit with relatively sparse data. 13,[19][20][21][22]49,50 The success of 5-ASA therapy has been variable, 49 These data emphasise the need for high-quality, prospective comparator studies to inform optimal management strategies for this emerging clinical problem.

Intervention and co-intervention
Outcome measures

Results and conclusions
Competing interest and source of support