The main inclusion criteria are as follows: 1) randomized controlled trials comparing ADA with conventional routine treatment (such as local and systemic corticosteroids, immunosuppressants) in patients of any age, gender, or ethnicity with a diagnosis of NIU; 2) the mean follow-up duration was more than three months; 3) sample size greater than 10; 4) AC cell grade and VH grade were evaluated by Standardization of Uveitis Nomenclature (SUN) criteria (Jabs, Nussenblatt, and Rosenbaum, 2005). We excluded studies that met any of the following conditions: 1) duplicate reports on the same study; 2) inadequate data or information; 3) control group was not placebo.

Pubmed, Embase, Web of Science, Cochrane Library, and Clinical Trials Registry were searched for relevant literature from their inception to November 19, 2020, regardless of language. If necessary, the researchers were contacted for more data. The search was limited to abstract/keyword/title fields. The search terms included uveitis, iridocyclitis, retinitis, retinal vasculitis, panuveitis, uveit*, adalimumab, ADA, Humira, TNF, TNF-a, anti-tumor necrosis factor-alpha, randomized controlled trial, and clinical trial. The Boolean operators appropriately connected these keywords. Study design types were restricted to randomized controlled trials (RCTs).

According to the inclusion criteria, two reviewers (Biao Li and Haoran Li) independently assessed the relevant studies for eligibility. Any disagreements were resolved through discussion among ourselves.

The primary outcome was the time to treatment failure (TTF), a rigorous composite outcome composed of four components (new ocular inflammatory lesions, BCVA, AC cell grade, and VH grade). “Treatment failure” was defined by the presence of one or more of the following factors: 1. new active, inflammatory lesions relative to baseline; 2. a two-step increase in anterior chamber cell or vitreous haze grade; 3. a worsening of best-corrected visual acuity by 15 or more letters on the Early Treatment Diabetic Retinopathy Study chart, relative to the best state previously achieved, in at least one eye; 4. sustained non-improvement with entry grade of ≥3; 5. use of concomitant medications not allowed; and 6. intermittent or continuous suspension of study treatment (adalimumab or placebo) for a cumulative period of longer than four weeks. The secondary efficacy outcomes included change in BCVA (logMAR), change in AC cell grade, and VH grade (according to SUN). The safety outcome was the number and the rate of adverse events (per patient-years).

Two authors (Biao Li and Haoran Li) extracted the data from the included publications into standard forms independently and cross-checked them to ensure accuracy. Differences were settled by discussion and transferred to a third author if needed. The information captured included: first author’s last name, published date, number of patients in each group, demographic data, follow-up time, and definitions of endpoints. If the same registered trial data appeared in multiple articles, the article with the latest or most comprehensive data was included.

Two reviewers (Biao Li and Li Zhang) independently evaluated the quality of the included studies using the recommended Cochrane Collaboration tool for assessing the risk of bias, which consists of seven types of risks of bias: random sequence generation; incomplete outcome data; allocation concealment; selective reporting; blinding of participants and personnel; blinding of outcome assessment; for-profit bias. Disagreements were settled by discussion and transferred to a third author if needed. We conduct our statistical analysis using review manager version 5.3 software according to the intention to treat analysis method. The evaluation of outcomes was done per eye, except TF and AEs, mainly pooled per patient. For continuous endpoints such as BCVA, AC cell grade, and VH grade, we preferentially retrieved mean differences with 95% CI. When meta-analysis was not proper for certain types of data we narratively summarize the relevant results.

The heterogeneity of the pooled results was assessed by Cochrane’s Q test and Higgins’ I². If apparent heterogeneity existed (p<0.1 or I² > 50%), the pooled results were estimated using the random-effects model. Alternatively, the fixed-effects model was adopted. Besides, we deleted each study to assess each study’s impact on the overall risk estimate to examine the results’ robustness. Subgroup analysis was initially planned according to the type of uveitis, study location, follow-up time duration, and participants’ age. Unfortunately, we do not have enough sample size to perform these analyses.

We originally planned to examine the publication bias of included studies by funnel plots and Egger’s test. Unfortunately, we lack enough studies to conduct these analyses.

The GRADE system was used to access the evidence’s quality of every efficacy outcomes. GRADE system scored the evidence of each outcome in five aspects: limitations of the study design and execution, inconsistency, indirectness, imprecision of results, and publication bias. Accordingly, we classify ADA treatment’s recommendation level as very low, low, medium, or high.

Our literature search yielded 918 articles (Pubmed: 335; Embase: 292; Web of Science: 161; Cochrane Library: 113; ClinicalTtrials.gov: 17). After removing duplicates, 616 articles remain. Of these, 568 were excluded after screening for the titles and abstracts. After the full-text examination of the remaining 48 articles, five articles (6 trials, there was one article containing the data of two RCTs) met inclusion criteria for our meta-analysis (Figure 1).

All six studies used intention-to-treat analysis to evaluate efficacy and safety outcomes. Six hundred thirty-six patients were enrolled in six studies, including 318 in the ADA group and 287 in the control group. The sample size ranged from 16 to 225. Overall, 61.23% (368/601) of patients were female, with the sex distribution favored females in each of the six studies. The mean age varied from 8.90 to 50.90 years old, and the mean uveitis duration was between 43.75 and 94.56 months. Table 1 summarises the characteristics of the six selected RCTs.

We judged for-profit bias in all six studies as high risk since five of the included studies were sponsored by pharmaceutical companies (AbbVie), while the remaining one studies had participants who received remuneration such as speaker’s fees from AbbVie.

Since only outcome assessors were blinded in Mackensen’s study, the risk of performance bias was considered high, and the detection bias was assessed as low risk. In the other five studies, the risk regarding blinding were all considered low risk since the methods of blinding patients, doctors, and outcome assessors were described in detail. The remaining four types of risk of bias were all considered low risk in six studies because the corresponding evidence can be found in the article. As prospective trial registrations were accomplished in all studies and their prespecified outcomes were reported, selective reporting was considered low risk. Figure 2 demonstrates the risk of bias in each study.

The pooled results included the time to treatment failure, change in AC cell grade, change in VH grade, and change in BCVA (logMAR). Table 2 demonstrates the relevant results of the including studies.

It cannot be performed due to the small sample size.

This meta-analysis of six RCTs including 605 patients systematically reviewed ADA’s efficacy and safety in NIU. The results show that ADA almost halved the risk of NIU patients’ treatment failure by significantly improving BCVA and reducing the AC cell grade and VH grade. The incidence of ADA-related AEs was generally low, and the safety profile was similar to other reports in previous studies. Significant differences favoring ADA over placebo was seen for two secondary endpoints (change in BCVA and VH grade). Outcomes regarding AC cell grade in the ADA group were numerically superior to that in the placebo group.

According to the GRADE, the certainty of the evidence concerning four efficacy outcomes were all judged as moderate. Table 4 shows the summary of GRADE’s. Moderate-quality evidence shows that ADA considerably lowered the risk of treatment failure or visual impairment, moderately reduced AC cell grades and VH grades in NIU.

The efficacy and safety results mentioned above were consistent with those of a meta-analysis by Shuai Ming et al. Their review included 20 non-RCTs and three RCTs. In contrast, our study included six RCTs and excluded non-RCTs, leading to a higher quality of evidence.

A similar article entitled “Anti-TNF Drugs for Chronic Uveitis in Adults—A Systematic Review and Meta-Analysis of Randomized Controlled Trials” has been published by Leal et al. on May 24, 2019. Leal’s study included three RCTs concerning two drugs: adalimumab and etanercept, with a sample size of 458. In comparison, our meta-analysis included six studies related to one single drug (ADA) with a sample size of 605. Moreover, the efficacy analysis was not conducted in Leal's study because of the significant heterogeneity between interventions. Our study pooled the six studies for efficacy analysis using uniform outcome measures such as BCVA, change in AC cell grade, and VH grade (SUN). This review is the first meta-analysis including only RCTs to summarize the current evidence regarding ADA’s efficacy and safety in NIU, with strengths in the relatively higher quality of evidence and multicomponent primary endpoint.

The principal limitation is that high heterogeneity was observed in our analysis. Differences in types of uveitis, patient’s age, concomitant medications, and outcome measures may contribute to the heterogeneity. It is important to note that we did not restrict participants’ age, so we recruited both adult and pediatric patients in this meta-analysis. In adults, ADA’s main indications in NIU are intermediate, posterior forms of uveitis and panuveitis. In children with uveitis, ADA’s main indication is JIA-associated uveitis and idiopathic uveitis, which is mainly anterior uveitis. We originally planned to conduct subgroup analysis in this article based on the location and type of uveitis, but due to the small sample size and the inability to extract relevant data from some studies, we gave up the subgroup analysis. In the future, more studies on the treatment of single uveitic disease with ADA are needed.

Besides, it is well known that RCT is not an ideal type of study to identify safety results because the relatively small sample size and short follow-up time make it challenging to identify rare adverse events. Therefore, the RCTs included in this review are not sufficient to study AEs thoroughly.

A further limitation was that the four VISUAL trials were sponsored by one pharmaceutical company (AbbVie) and the remaining two studies had participants who received remuneration such as the speaker’s fees from AbbVie, which may seriously affect the results.

In the future, independent non-company sponsored RCTs are needed to further provide more objective and robust evidence. Besides, it is necessary to further compare ADA with conventional immunosuppressors regarding efficacy and safety in NIU.