Amniotic membrane transplantation for infectious keratitis: a systematic review and meta-analysis

Infectious keratitis (IK) is the 5th leading cause of blindness globally. Broad-spectrum topical antimicrobial treatment is the current mainstay of treatment for IK, though adjuvant treatment or surgeries are often required in refractory cases of IK. This systematic review aimed to examine the effectiveness and safety of adjuvant amniotic membrane transplantation (AMT) for treating IK. Electronic databases, including MEDLINE, EMBASE and Cochrane Central, were searched for relevant articles. All clinical studies, including randomized controlled trials (RCTs), non-randomized controlled studies and case series (n > 5), were included. Primary outcome measure was time to complete corneal healing and secondary outcome measures included corrected-distance-visual-acuity (CDVA), uncorrected-distance-visual-acuity (UDVA), corneal vascularization and adverse events. A total of twenty-eight studies (including four RCTs) with 861 eyes were included. When compared to standard antimicrobial treatment alone, adjuvant AMT resulted in shorter mean time to complete corneal healing (− 4.08 days; 95% CI − 6.27 to − 1.88; p < 0.001) and better UDVA (− 0.26 logMAR; − 0.50 to − 0.02; p = 0.04) at 1 month follow-up in moderate-to-severe bacterial and fungal keratitis, with no significant difference in the risk of adverse events (risk ratio 0.80; 0.46–1.38; p = 0.42). One RCT demonstrated that adjuvant AMT resulted in better CDVA and less corneal vascularization at 6 months follow-up (both p < 0.001). None of the RCTs examined the use of adjuvant AMT in herpetic or Acanthamoeba keratitis, though the benefit was supported by a number of case series. In conclusion, AMT serves as a useful adjuvant therapy in improving corneal healing and visual outcome in bacterial and fungal keratitis (low-quality evidence). Further adequately powered, high-quality RCTs are required to ascertain its therapeutic potential, particularly for herpetic and Acanthamoeba keratitis. Future standardization of the core outcome set in IK-related trials would be invaluable.

To date, a number of studies have evaluated the benefit of AMT for treating active IK, though the majority of them were of small case series or case reports. In clinical practice, AMT is usually reserved as a second-line therapy in IK, mainly to promote cornea healing in non-healing ulcer after the sterilization phase. Therefore, the value of employing AMT in addition to standard antimicrobial treatment (SAT) during the active phase of IK remains uncertain. Liu et al 24 recently conducted a systematic review of 17 studies on the use of AMT for infective and non-infective corneal ulcers. Although the review provided a detailed analysis on the corneal healing rate and visual improvement rate associated with the use of AMT, it did not compare the effect of adjuvant AMT and SAT with SAT alone. Furthermore, the study did not include several important relevant studies, including three randomized controlled trials (RCTs) [25][26][27] and one non-randomized controlled study (NRCS) 28 , rendering the robustness of evidence uncertain. In light of these limitations, we conducted a systematic review and meta-analysis to critically examine the effectiveness and safety of combined AMT and SAT versus SAT alone in treating patients with IK.

Results
Literature search and study characteristics. The electronic searches last conducted on 01 November 2020 retrieved a total of 969 titles and abstracts (see Fig. 1 for the PRISMA flow chart). After removing 262 duplicates and including three additional records identified through other sources, the remaining 709 records were screened and 669 references that were not relevant to the scope of the review were excluded. A total of 35 Meta-analysis of eligible RCTs. Overall description. Four eligible RCTs were included in the metaanalysis, which included a total of 209 eyes that compared the effectiveness of SAT and adjuvant AMT with SAT alone [25][26][27]29 . These consisted of four single-centre RCTs conducted separately in China (n = 2) 27,29 , India (n = 1) 26 , and Iran (n = 1) 25 . Two additional RCTs were also identified but were excluded from the meta-analysis as one RCT compared AMT with conjunctival flap 53 and another RCT compared AMT and argon laser treatment with AMT for treating IK 54 . There was no ongoing trial identified from the clinical trial registry databases. The RCTs included participants with an average age between 45.5 and 55.6 years (ranged 15-98 years) with a slight (61.7%) male preponderance. None of the RCTs were prospectively registered with any clinical trial database.
Types of microbes and severity of IK. The four RCTs (n = 209 eyes) included a total of 108 (51.7%) eyes with bacterial keratitis, 95 (45.5%) eyes with fungal keratitis and 6 (2.9%) eyes with mixed bacterial/fungal keratitis. No herpetic or Acanthamoeba keratitis was included in any of the RCTs. Li et al. 29 and Zeng et al. 27 included only fungal keratitis and Tabatabaei et al. 25 included only bacterial keratitis whereas Arya et al. 26 included a mixed cohort, which included bacterial, fungal or mixed bacterial/fungal keratitis. The proportion of the types of microorganisms was similar between the treatment and control groups among all RCTs.
The baseline reporting and extent of the severity of IK were heterogeneous among the four RCTs. Two studies reported the baseline ulcer diameter or area in continuous values 25,27 , one study reported the baseline severity in categorical values (i.e. 1-2 mm, 2-5 mm, or > 5 mm in diameter) 26 , and one study did not report the baseline severity of IK 29 . Tabatabaei et al. 25 and Zeng et al. 27 included IK cases with a mean ulcer diameter of 5.2 mm and 6.7 mm, respectively (i.e. severe IK based on previous studies) 55,56 . Similarly, Arya et al. 26 included mainly moderate-to-severe IK, with 47.5% of moderate ulcer (2-5 mm in diameter) and 42.5% of severe ulcer (> 5 mm in diameter). None of the RCTs included eyes presented with threatened or actual corneal perforation.
Surgical technique and timing of AMT. In terms of the preservation technique, two RCTs utilized cryopreserved AMs 26,27 , one RCT utilized freeze-dried AMs 29 , and one did not report the type of AMs used. Three RCTs 26,27,29 employed AMT as an overlay patch (single-or double-layer) whereas one RCT 25 employed AMT as an inlay graft (double-layer). All AMTs were performed with the epithelium/basement membrane side up. The time interval between the initial presentation of IK and AMT was 3 days in one study 29 , 2-5 days in one study 25 , and unspecified in two studies 26,27 .
Risk of bias. The risk of bias of the RCTs, based on the five abovementioned domains, is summarized in Fig. 2. Only one RCT 25 clearly reported the randomization process, with details on randomization and allocation concealment. All RCTs had complete or nearly complete (> 99%) follow-up data. There were some concerns about the measurement of the outcome as it was not possible to mask the outcome assessor (i.e. AMT would be visible on postoperative clinical examination, hence revealing the intervention arm). In addition, there were some concerns about selective reporting of outcomes across all RCTs as none of them were prospectively registered with any publicly accessible clinical trial database; therefore, the risk of bias could not be confidently evaluated. No relevant conflict of interest was declared in any of the RCTs.
The risk of bias of NRCS was assessed using the ROBIN-I tool. Two NRCS 28,31 were judged to be at serious risk of bias due to bias in selection of participants and measurement of outcomes. One NRCS 30 was judged to be at critical risk of bias due to baseline confounding, bias in selection of participants and measurement of outcomes.
Effects of intervention. The effects of interventions were categorized into: (A) time to complete corneal healing (defined by complete corneal re-epithelialization and resolution of infection); (B) UDVA and CDVA (in logMAR) at 1-6 months follow-up; (C) size of ulcer or infiltrate; (D) extent of corneal neovascularization at 1-6 months; and (E) adverse events defined by worsening IK, endophthalmitis or corneal perforation requiring corneal gluing, tectonic keratoplasty or evisceration during the follow-up period. The GRADE summary of findings for each treatment outcome is summarized in Table 3.
Time to complete corneal healing. Three RCTs (n = 169 eyes) reported the primary outcome measure, which was the time to complete corneal healing 25,27,29 . There is very low-quality evidence that adjuvant AMT expedited the time to complete corneal healing compared to SAT alone [mean difference (MD) − 4.08 days; 95% CI − 6.27 to − 1.88; I 2 = 18%; p < 0.001; Fig. 3A]. The quality of evidence was downgraded due to the high risk of bias and imprecision. The randomization process was not clear in one RCT 27 and allocation concealment were not performed in two RCTs 27,29 . Furthermore, there was lack of blinding of the participants and the assessors across all  25,27,29 . In addition, the total number of participants/eyes pooled in the meta-analysis was less than the number generated by a conventional sample size calculation.
Uncorrected-distance-visual-acuity (UDVA). Two RCTs (n = 119 eyes) reported the UDVA at 1 month followup 25,27 . There is very low-quality evidence that adjuvant AMT resulted in better UDVA compared to SAT alone (MD − 0.26 logMAR; − 0.50 to − 0.02; I 2 = 0; p = 0.04; Fig. 3B). The quality of evidence was downgraded due to high risk of bias and imprecision. The randomization process and allocation concealment were not clear in one RCT. There was lack of blinding of the participants and assessors and the sample size was limited in both RCTs. Only one RCT (n = 99 eyes) reported the UDVA at 6 months follow-up, where the adjuvant AMT group achieved a better UDVA than the SAT alone group (1.34 ± 0.69 vs. 1.69 ± 0.54; p < 0.001) 25 .
Corrected-distance-visual-acuity (CDVA). Only one RCT (n = 99 eyes) reported the CDVA at 1-6 months follow-up, therefore meta-analysis was not possible 25  Size of ulcer or infiltrate. We did not identify any RCT that reported the size of ulcer or infiltrate. However, one RCT (n = 99 eyes) reported the area of corneal scarring (in mm 2 ) at 6 months follow-up, which was found to better in the adjuvant AMT group than the SAT alone group (17.6 ± 5.7 vs. 22.8 ± 6.1; p < 0.001) 25 . The quality of evidence was downgraded due to some risk of bias (lack of blinding) and imprecision (limited sample size).
Corneal vascularization. Only one RCT (n = 99 eyes) reported the extent of corneal vascularization, measured as % of the total area of corneal surface, at 1-6 months follow-up 25 . There is low-quality evidence that adjuvant AMT resulted in less corneal vascularization (%) at 1 month (5.0 ± 4.0 vs. 7.0 ± 5.0; p < 0.001) and 6 months follow-up (2.0 ± 3.0 vs. 7.0 ± 6.0; p < 0.001). The quality of evidence was downgraded due to some risk of bias (lack of blinding) and imprecision (limited sample size).
Adverse events. All four RCTs (n = 209 eyes) reported the occurrence/risk of adverse events, defined as worsening of IK or perforation requiring corneal gluing or tectonic keratoplasty [25][26][27]29 . There is very low-quality evidence that there was no evidence of difference in the risk of adverse events between adjuvant AMT and SAT alone groups at 1-6 months follow-up (RR 0.80; 95% CI 0.46-1.38; I 2 = 0; p = 0.42; Fig. 3C). The risk of adverse events was noted to be considerably variable across four RCTs, with a risk ranging between 0.0 and 26.5% in www.nature.com/scientificreports/ the AMT group and 10.0-32.0% in the SAT group. Such heterogeneity was likely related to the variations in the patient cohort, types and severity of IK, and the standard treatment regimen employed among different studies. For each specific complication, adjuvant AMT group had a slightly lower risk than the SAT alone group in terms of worsening of IK requiring secondary debridement (0-10% vs. 0-20%), secondary glaucoma (4% vs. 36%), and corneal perforation requiring either corneal gluing or tectonic keratoplasty (0-26.5% vs. 10-32%). The quality of evidence was downgraded due to high risk of bias and imprecision. The follow-up duration was variable across studies, ranging from 1 to 6 months. One RCT (n = 50 eyes with fungal keratitis) reported the risk of secondary glaucoma, which was 4% (n = 1 eye) in the AMT group and 36% (n = 9 eyes) in the SAT group (p = 0.011) 30 .
According to the CONSORT reporting of harms guidance, all four RCTs were judged to be of inadequate quality as the severity of adverse events and clinical sequalae were not clearly described [25][26][27]29 .
Subgroup analysis based on the type of IK. There were insufficient RCTs to perform subgroup analysis between bacterial and fungal keratitis outcomes. In view of the limitation, pooled estimates on time to complete corneal healing and adverse events were assessed based on RCTs and NCRSs. Pooled estimates of the time to complete corneal healing showed that adjuvant AMT expedited the time to complete corneal healing in bacterial keratitis (MD − 2.42 days; 95% CI − 4.53 to − 0.32; p = 0.02) and fungal keratitis (MD − 6.90 days; 95% CI − 11.58 to − 2.21; p = 0.004; Table 4). In terms of the risk of adverse events, the pooled estimates demonstrated that adjuvant AMT did not significantly influence the risk of adverse events in bacterial keratitis (RR 0.83; 95% CI 0.45-1.54), fungal keratitis (RR 0.28; 0.05-1.65), and mixed bacterial/fungal keratitis cohorts (RR 1.50; 0.28-8.04; Table 4). Subgroup analysis of the outcomes based on large non-comparative case series (n ≥ 5 eyes) is summarized in Table 5. Studies that reported the pooled results with no distinction made to the underlying microorganisms were excluded from the subgroup analysis. Based on these case series, complete corneal healing rate was 92.6% (113/122) in bacterial keratitis, 80.0% (96/120) in fungal keratitis, 93.8% (137/146) in herpetic keratitis, in 77.8% (7/9) Acanthamoeba keratitis, 81.8% (9/11) in mixed bacterial and fungal infection, and 100% (4/4) in mixed bacterial and herpetic infection.

Discussion
To the best of our knowledge, this study represents the most up-to-date systematic review and meta-analysis examining the effectiveness and safety of adjuvant AMT for treating IK. The use of AMT for ocular diseases dates back to the early twentieth century 57 . However, the procedure was not commonly performed until 1990s where Batlle and Perdomo demonstrated the wound healing property of AM, reinvigorating the interest of utilising AMT for ocular surface disease 8 . The use of AMT for ocular surface reconstruction was subsequently popularized by Kim and Tseng 58 using glycerin-preserved AM. Although AMT has since been performed to treat a wide array of ocular surface diseases, high-quality evidence of using AMT in treating active IK remains limited. So far, there has only been one systematic review that had partially examined the benefit of AMT in IK 24 . The systematic review included 17 studies (n = 390 eyes) and examined the healing rate and visual improvement rate after AMT in either infectious or non-infectious corneal ulcers. While the review showed that AMT was effective in treating IK, it did not answer a very important clinical question, which is whether adjuvant AMT provides any additional benefit or risk during the management of IK, in addition to standard antimicrobial treatment. In addition, the review did not capture 3 RCTs that were identified in our systematic review, rendering the evidence of their findings uncertain. Furthermore, the effect of AMT has not been examined in the context of the types of organisms.
Summary of main findings. In this systematic review, we included a total of 28 studies (n = 861 eyes), encompassing four RCTs with 209 eyes and 24 non-RCTs with 657 eyes. Among the RCTs, the majority of the included eyes were either affected by bacterial keratitis or fungal keratitis, with only 6 eyes being affected by mixed bacterial and fungal keratitis. None of the RCTs examined the effect of adjuvant AMT in herpetic or Acanthamoeba keratitis. Based on the meta-analysis of three RCTs, we demonstrated that early adjuvant AMT shortened the time to complete corneal healing by approximately 4 days when compared to SAT alone. In addition, it was shown that IK patients treated with adjuvant AMT achieved 0.26 logMAR (equivalent to 2-3 Snellen lines) better UDVA at 1 month follow-up compared to SAT alone. No evidence of difference was noted in terms of the risk of adverse events. Furthermore, other beneficial effects such as better CDVA and less corneal vascularization were shown in the adjuvant AMT group.
The observed beneficial effects in the adjuvant AMT group are likely attributed to the dual antimicrobial and anti-inflammatory properties of AMT 8,16 . The plausible mechanisms of antimicrobial effect of AMT in IK are at least twofold. First, the antimicrobial activity is directly linked to the presence of various antimicrobial components, including lysozyme, transferrin, and immunoglobulin, in the amniotic fluid 8,59 . Second, studies have shown that AM could serve as an effective antibiotic reservoir when used in combination with antibiotics and provide sustained drug delivery 60 . Furthermore, AM has been shown to exhibit anti-inflammatory function via regulation of T-cell function and secretion of anti-inflammatory antagonists, including IL-1ra, sTNF, and VEGF-R 61,62 . This anti-inflammatory property will have a long-term beneficial effect on the cornea, hence vision, as persistent corneal vascularization could negatively affect the vision, either directly via encroachment on the visual axis or indirectly via lipid keratopathy where exudation of lipid and inflammatory cells extends to the visual axis 63 . Moreover, it increases the risk of graft rejection should corneal transplantation needs to be carried out to restore the optical clarity of the cornea 63 www.nature.com/scientificreports/ Although the benefit of AMT in herpetic keratitis has not been ascertained in RCTs, our systematic review (based on large case series) showed that it enabled a high rate (94%) of complete corneal healing. Herpetic keratitis, particularly the necrotizing form, is a potentially sight-threatening condition that is difficult to treat clinically 65 . It is also notoriously known to be associated with neurotrophic keratopathy, which results in delay in corneal healing 65,66 . Therefore, AMT serves as a useful adjunct treatment in this clinical circumstance in view of its dual anti-viral and wound healing properties 67,68 . On the other hand, only two case series have reported the use of AMT in Acanthamoeba keratitis, albeit good effect was observed. The broad-spectrum antimicrobial activity of AMT (observed in our systematic review) against a wide range of microorganisms is particularly beneficial in the management of IK as polymicrobial keratitis is a relatively common entity, which often poses significant diagnostic and therapeutic challenges 4,5,69 . Based on large case series, adjuvant AMT has been shown in our review to be an effective treatment for mixed bacteria/fungal keratitis (82% complete healing; n = 9/11) and mixed bacterial/herpetic keratitis (100% complete healing; n = 4/4).
A number of studies have also demonstrated that AMT, when employed as multi-layer, could effectively treat IK with threatened or actual corneal perforation (up to 5 mm) 35,37,43,52 . Studies have demonstrated that multilayer AMs can promote re-epithelialization of the cornea and integrate with the corneal stroma, with progressive repopulation of AM by cornea stroma-derived cells, ultimately resulting in corneal healing and thickening 70 . This approach helps obviate the need for an emergency therapeutic keratoplasty, which is known to be associated with high risk of graft failure (in the setting of active inflammation) and recurrence of infection 69,71,72 . It would also help reduce the burden on the availability of donor corneas, which have been significantly affected by the recent COVID-19 pandemic 73 .
Overall completeness and applicability of evidence. This study represents the most up-to-date systematic review and meta-analysis specifically examined the effectiveness and safety of adjuvant AMT in managing all types of IK. The four RCTs included primarily bacterial and fungal keratitis cases, therefore the outcome of the meta-analysis should be interpreted in the context of these types of infection. All RCTs were completed without any significant drop out and there was no concern with the safety of AMT.
The reporting and categorization of the baseline severity of IK was considerably variable among the RCTs. Two studies included corneal ulcers with a mean diameter of 5-7 mm (severe disease) 25,27 , one study included mainly moderate (2-5 mm in diameter) and severe (> 5 mm in diameter) ulcers, and one did not report the baseline severity. Such heterogeneity highlights the need for standardized reporting and a core outcome set in the future clinical trials related to IK. Of the three RCTs that reported the baseline severity of IK, the majority of the included participants/eyes were affected by moderate-to-severe IK, suggesting that adjuvant AMT would be useful for this group of patients. Future studies examining the effectiveness and safety of adjuvant AMT in mild-to-moderate IK would be required before being routinely applied in clinical practice.
Another aspect that was not examined in this systematic review is the effects of different surgical techniques and types of AMT on the clinical outcomes. In clinical practice, a number of AMT techniques have been described. It can be used as a patch (or overlay) where an AM that is transplanted to cover the ocular surface, allowing the host corneal epithelium to regenerate under the AM. It can also be used as graft (or inlay) where a smaller AM is grafted over the corneal epithelial defect and act as a substrate for the host epithelium to grow over. Occasionally, a sandwich AMT technique, utilizing a mixture of overlay and inlay, is performed to facilitate the regeneration of host epithelium between the AMs 16 . In addition, various forms of AMs, either fresh form, cryopreserved, lyophilized (freeze-dried), or air-dried AM, have been used, with comparable clinical efficacy observed amongst them 16 . While studies have shown the varying effects of preservation method on the structural and functional properties of AMs, strong clinical evidence is lacking in the literature. Cryopreserved AMs, depending on the duration of the preservation, have been shown to possess no or less viable cells compared to fresh, freeze-dried or vacuum-dried AM, which have more viable cells with proliferative capability 16,74 . On the other hand, some studies have demonstrated that lyophilization of the AMs may lead to a greater reduction in the growth factors compared to cryopreservation. Preclinical studies observed similar efficacy in terms of healing activity between cryopreserved and freeze-dried AMs, supporting the use of both types of AMs in clinical setting 75,76 . In our systematic review, the three RCTs included in the meta-analysis for time to complete healing  www.nature.com/scientificreports/ outcome, one RCT employed cryopreserved AMT, another RCT utilized freeze-dried AMT and the remainder did not report the preservation method used. Therefore, it is not possible to perform a head-to-head comparison between each type of AM due to limited studies, but the overall effect of either type was positive.
Quality of evidence and potential biases in the review. Similar outcome reporting in the included RCTs has enabled the meta-analysis, which demonstrated the value of adjuvant AMT in expediting complete  Table 3. GRADE summary of findings for various treatment outcomes of adjuvant amniotic membrane transplantation for infectious keratitis. GRADE Working Group grades of evidence. High certainty: We are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: We are moderately confident in the effect estimate: The true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: Our confidence in the effect estimate is limited: The true effect may be substantially different from the estimate of the effect. Very low certainty: We have very little confidence in the effect estimate: The true effect is likely to be substantially different from the estimate of effect. Explanations. a High risk of bias due to lack of randomization and allocation concealment in ≥ 50% of the included studies. b Potential risk of bias due to the lack of blinding in participants and assessors. c The total number of participants is less than the number generated by a conventional sample size calculation. d There are few events and the confidence interval includes appreciable benefit and harm. e Differences in final follow-up duration. CI confidence interval, MD mean difference, RR risk ratio, UDVA uncorrecteddistance-visual-acuity, CDVA corrected-distance-visual-acuity. *The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI).

Patient or population: Infectious keratitis Intervention: Adjuvant amniotic membrane transplantation (AMT) Comparison: Standard antimicrobial treatment (SAT)
Outcomes Anticipated absolute effects* (95% CI) www.nature.com/scientificreports/ corneal healing, reducing corneal vascularization and potentially achieving better visual outcome. However, two RCTs have some concerns on the risk of bias and another two RCTs have high risk of bias. Due to the lack of information in the randomization process, lack of blinding, and under-powered sample size. In addition, none of the RCTs were prospectively registered with any clinical trial database, which limits the assessment of selective reporting of the outcomes. Of all four RCTs, only Tabatabaei et al. 25 conducted a sample size calculation based on the anticipated difference in the size of corneal scar (mean effect size of 4 mm 2 , standard deviation of 6 mm 2 , and power of 90%), which yielded a sample size of 48 eyes in each group. However, it is noteworthy to mention that size of corneal scar is not a routine outcome measure for IK. Based on the findings of our systematic review, an appropriate sample size (using time to complete corneal healing as the main outcome measure) is estimated to be at approximately 200 participants/eyes (based on a mean difference or effect size of 4.08 days, standard deviation of 14.56 days, power of 80% and alpha of 5%). This sample size is also supported by some previous RCTs examining the effectiveness of therapeutic CXL (or PACK-CXL) for IK, where an appropriate sample size was estimated at around 200-250 participants, using time to complete corneal healing as the main outcome measure 10,77 . This large sample size would then potentially control for various potential confounding variables that could affect the time to complete healing, including the demographic factors, initial size and severity of the corneal ulcer, types of organisms, and underlying comorbidity (e.g. diabetes).   Table 5. Summary of the healing rate and treatment failure of amniotic membrane transplantation (AMT) in large case series (N > 5 eyes) based on the types of causative microorganisms.

Authors
Year www.nature.com/scientificreports/ Another important bias, which is a common issue in surgical trial, is related to the difficulty in blinding (or masking) the participants and the surgeons 78,79 . This is evident in our systematic review where none of the RCTs had implemented or reported any measure to ensure the masking of participants or surgeons, which could potentially lead to risk of bias. In a trial, masking can be introduced to 5 groups of individuals, including participants, surgeons, data collector, outcome adjudicators and data analysts. While it is always possible to mask the latter three groups (i.e. data collectors, outcome adjudicators and data analysts), masking of the participants or surgeons may not always be possible, depending on the surgical intervention and design. For instance, in AMT trial, it would have been difficult to mask the patients or surgeons as the patients would have known that an AMT has been performed (unless a sham surgery is performed) and the surgeons would have seen the physical presence of the AM postoperatively. However, masking the other 3 groups of individuals could reduce the bias. As fewer than 10 studies were eligible for inclusion, we were unable to use a funnel plot to identify possible publication bias.
In conclusion, this systematic review and meta-analysis demonstrated the benefit of early adjuvant AMT in accelerating corneal healing and improving visual outcome in moderate-to-severe bacterial and fungal keratitis. However, further adequately powered and well-designed RCTs are required to ascertain the true potential of adjuvant AMT in treating active IK, particularly herpetic and Acanthamoeba keratitis. Future standardization of the baseline assessment and core outcome set for clinical trials related to IK would also be invaluable. who. int/ ictrp) for primary research related to AMT for IK. There was no date or language restriction in the search for trials. Electronic databases were first searched on 01 March 2020, followed by a final update on 01 November 2020. The bibliographies of the included articles were reviewed for any additional eligible articles. Key words used were "amnion", "amniotic membrane", "corneal ulcer", "corneal infection", "infectious keratitis", and "microbial keratitis". Search strategies for MEDLINE and EMBASE are provided in the Supplemental Table S1.

Study selection.
All clinical studies, encompassing RCTs, NRCSs and case series (n ≥ 5 eyes), related to AMT for IK were included as few RCTs were anticipated. The analysis was conducted at two levels; (1) a metaanalysis of all eligible RCTs and (2) a systematic review of all clinical studies, including NRCSs and case series. All types of IK, including bacterial, fungal, viral, parasitic or mixed infection, were included in this review. Studies that evaluated cases of non-infectious keratitis, other types of surgical interventions, or AMT used for nonantimicrobial purpose were excluded from this study. Small case series (N < 5 eyes), reviews, published abstracts, laboratory and animal studies were also excluded. For the meta-analysis, the intervention group included cases of IK that were treated by AMT and SAT whereas the control group included cases of IK that were treated with SAT alone. There was no restriction applied to the published language, location or setting of the study, or patient demographic factors. This study conformed to the Preferred Reporting Items for Systematic reviews and Meta-Analysis (PRISMA) guideline (see Supplementary Table S2) 81 .

Data extraction.
A web application designed for systematic reviews, Rayyan (Qatar) 82 , was used to help collate the potential studies and expedite the initial screening of abstracts and titles. The titles and abstracts obtained from the searches were independently screened by two authors (D.S.J.T. and C.H.) to include studies that fulfilled the eligibility criteria. The authors then independently assessed the full-text version of all the selected articles and extracted data onto a standardized data collection form for qualitative review. The extracted data included the authors, year of publication, sample size, types of interventions, types of causative microorganisms, outcomes and complications 80 . Discrepancies were resolved by group consensus if consensus could not be reached.
For the meta-analysis, the following information were extracted from the included RCTs and entered into RevMan (Review Manager 5.4) software 80 www.nature.com/scientificreports/ Outcome measures. For the meta-analysis, the primary outcome measure was the time to complete corneal healing (defined as complete corneal re-epithelialization and clearance of infiltrate and hypopyon; days) and the secondary outcome measures included the corrected-distance-visual-acuity (CDVA; logMAR) and uncorrected-distance-visual-acuity (UDVA; logMAR) at final follow-up (1-6 months), size of corneal ulcer (at 1-6 months), extent of corneal vascularization, and risk of adverse events (defined as worsening IK and/or corneal melt/perforation requiring gluing or tectonic/therapeutic keratoplasty or evisceration) at final follow-up (1-6 months). These outcome measures were similarly used in our previous systematic review and meta-analysis in examining the effectiveness and safety of adjuvant PACK-CXL for IK 10 . Relevant data of all included studies were summarized and reported. Continuous variables such as time to complete corneal healing, UDVA, CDVA, and size of corneal epithelial defect and infiltrate, were presented as mean with standard deviation (SD). In studies that reported median and interquartile range, the means and SDs were estimated using formulas reported by Wan et al. 84 and the Cochrane Handbook estimator 85 . Dichotomous variable such as risk of adverse events was defined by the number of eyes with adverse events.

Assessment of risk of bias.
Risk of bias was assessed by two authors (D.S.J.T. and C.H.) independently and any disagreement was adjudicated by group consensus. Included RCTs were assessed for sources of systematic bias using the revised Cochrane risk of bias tool (RoB 2 tool) 86 . The review authors were not masked to the authors of the studies during this assessment. A judgement of risk of bias as 'high risk' , 'some concern' , or 'low risk' was made for the following domains: (1) randomization process; (2) deviations from intended interventions; (3) missing outcome data; (4) measurement of the outcome; and (5) selection of the reported result 86 . NRCSs were assessed for risk of bias using the ROBINS-I tool 87 against seven domains; the worst judgement in any of the domains was used as the overall risk of bias.
Measure of treatment effect. Dichotomous data were measured as risk ratios (RRs) with 95% confidence intervals (CI) and continuous data as mean differences (MDs) with 95% CI 80,88 . The unit of analysis was the eye as the patient might have different causes of infection and treatment in each eye. There was no issue with the unit of analysis in the included RCTs. The review was conducted based on the available data from the trials. When data were unavailable but the level of missing data and reasons for missing data in each group were similar, data were analyzed even when intention-to-treat (ITT) analysis was not performed.
Assessment of heterogeneity. The heterogeneity of the RCTs and NRCSs was checked by careful review of the full-text, assessment of forest plots and examination of the I 2 value with its confidence interval. The overall characteristics of the studies, in particular the types of participants, causes of IK and types of interventions were examined to assess the extent to which the studies were similar enough to make pooling study results sensible 80 . The results of forest plots were reviewed for consistency of the size and direction of effects. I 2 values greater than 50% were considered indicative of substantial heterogeneity and meta-analysis could not be conducted due to inconsistency of effect estimates 85 . Random-effects model was used for the meta-analysis as some degree of heterogeneity will always exist due to clinical and methodological differences of the studies 80 . The Chi 2 p value was also considered as this has a low power when the number of studies were few. A p value of < 0.1 was considered statistically significant 85 . Data synthesis and analysis. A meta-analysis was undertaken when there were sufficient similarities in the reporting of outcome measures 80 . A random-effects model in RevMan 5.4 was used in view of the expected heterogeneity across different studies. The Mantel-Haenszel method was employed for analysing the risk ratio (RR) of adverse events in view of the small expected number of events 80 . If there was inconsistency between the results of individual studies such that a pooled result might not be a good summary of the individual trial results-for example, the effects were in different directions or I 2 > 50% and p < 0.1-the data were not pooled but described in narrative format 80,88 . Where there was statistical heterogeneity the data were pooled when all the effect estimates were in the same direction, such that a pooled estimate would seem to provide a good summary of the individual trial results. Sensitivity analysis was performed by assessing the impact of including studies at high risk of bias for an outcome in one or more key domains 80 . This was conducted by omitting each study in turn to examine the influence of individual studies (with high risk of bias) on the overall pooled estimate. A summary of findings is presented below including the assessment of the quality of the evidence for outcomes using the GRADE approach with GRADE Pro/GDT software 89 . All RCTs were started with a rating of 'high-quality' evidence and were downgraded by one level for serious concerns (or by two levels for very serious concerns) regarding the risk of bias, inconsistency, indirectness, imprecision and publication bias. The quality of evidence of studies was graded by two assessors (D.S.J.T and C.H.) independently and any disagreement was adjudicated by group consensus.
Subgroup analysis based on the type of organisms. In addition to the meta-analysis, a subgroup analysis of the effectiveness and adverse events based on different types of organisms was performed. In view of the anticipated low number of RCTs, both experimental and quasi-experimental study designs including RCTs and NRCSs were included in the subgroup analysis. Descriptive case series were also reviewed for rare or uncommon adverse events. Pooled estimates of the time to complete corneal healing and risk of adverse events across comparative studies, including RCTs and NRCSs, were calculated. Causes of IK were categorized as bacterial, fungal, viral, Acanthamoeba, or mixed infection.