Pars Plana Vitrectomy with or without Internal Limiting Membrane Peel for Epiretinal Membrane: A Systematic Review and Meta-Analysis

Epiretinal membrane (ERM) is characterized by a proliferation of transdifferentiated Müller cells and myofibroblasts associated with the extracellular matrix that overlies the internal limiting membrane (ILM) scaffold, leading to decreased central vision and metamorphopsia [1, 2]. Surgical removal of ERM typically leads to a significant reduction in retinal thickening and architecture, which may be more effective when accompanied by ILM peeling [3]. At the same time, the importance of ILM peeling during pars plana vitrectomy (PPV) remains widely debated for patients with ERM [4]. Although ILM peeling increases retinal compliance and may result in better anatomic outcomes, it also removes Muller cell footplates which can lead to Muller cell dysfunction [5].

Adjuvant dyes or agents such as indocyanine green (ICG), brilliant blue (BB) and trypan blue (TP), or triamcinolone acetonide (TA) are typically used to increase contrast during ILM peeling [6]. Although staining of the ILM with adjuvants may facilitate peeling, it may also result in retinal toxicity [4, 6]. Specifically, ICG has photosensitizing properties which may induce inner retinal toxic reactions following ILM peeling [6]. ILM peeling may also lead to iatrogenic damage of the nerve fiber layer and the macula, which may affect visual function [6‒8].

Overall, the safety and effectiveness of ILM peeling during PPV remains poorly understood for the treatment of ERM. Our study aims to provide a comprehensive update on the comparative safety and effectiveness of PPV with and without ILM peeling for ERM.

We conducted a systematic literature search on Ovid MEDLINE, EMBASE, Cochrane Library, and Google Scholar from January 2000 to January 2023. A summary of our search strategy on Ovid MEDLINE can be found in online supplementary eTable 1 (for all online suppl. material, see https://doi.org/10.1159/000534851). We included randomized controlled trials (RCTs) and observational studies comparing visual and anatomical outcomes between patients receiving PPV with or without ILM peeling for ERM. We excluded (1) studies that did not present outcomes stratified by whether the ILM was peeled during PPV, (2) studies on inverted ILM flap surgery, (3) studies in which some patients had retinal detachment prior to surgery, (4) studies where data from idiopathic and secondary ERM patients could not be distinguished, and (5) studies that did not provide applicable data for at least one of our primary endpoints. Studies with no English full-text or no published results, systematic reviews, meta-analyses, literature reviews, cost-benefit analyses, conference abstracts, theses, and letters to editors were excluded. Our study did not require institutional ethics approval by virtue of its design, and we adhered to the Declaration of Helsinki. Our protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO) with ID CRD42022296360.

We performed title and abstract screening followed by full-text screening on Covidence (Veritas Health Innovation, Melbourne, Australia). We collected all data from included studies on Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA). At least two independent authors (A.M., H.A., N.P.) performed screening and study selection and at least two independent authors collected data (A.M., R.H.). A third impartial author (M.M.P.) was consulted for conflict resolution as necessary. We collected the following baseline characteristics: country, year of publication, proportion of males, number of eyes, proportion of phakic eyes, proportion of pseudophakic or aphakic eyes, number of right eyes, condition, treatment, initial best-corrected visual acuity (BCVA) in logMAR, initial retinal thickness (RT), and type of RT reported. We considered any of the following RT outcomes for collection: central retinal thickness, central foveal thickness, central subfield foveal thickness, and central macular thickness (CMT). Our primary endpoints were BCVA at last study observation and change in BCVA from baseline. We excluded studies that did not provide the mean, standard deviation (SD), or sample size of at least one primary outcome. Our secondary endpoints were RT at last study observation, change in RT from baseline, ERM recurrence with or without additional surgery, and other adverse events. We contacted authors by electronic mail to request missing data and sent follow-up emails 2 weeks later if there was no initial response. When two or more studies reported on the same patient dataset, the study accepted most recently to a peer-reviewed journal was included and previous versions were consulted for missing data.

Two independent reviewers evaluated the risk of bias of included studies and certainty of evidence of study outcomes (R.H., H.A.). A third independent author (M.M.P.) was consulted for conflict resolution as necessary. We used the following Cochrane tools to evaluate the risk of bias of included studies: Risk of Bias 2 (ROB2) tool for RCTs [9] and Risk of Bias in Non-randomized Studies of Interventions (ROBINS-I) tool for observational studies [10]. We used the Grading of Recommendations, Assessment, Development and Evaluation (GRADE) tool to evaluate the certainty of evidence of each outcome [11].

We reported weighted mean differences (WMDs) and SDs for continuous outcomes. We reported risk ratios (RRs) and the number needed to treat for discrete outcomes. We defined the mean as $x¯=∑i=1nwixi∑i=1nwi$ and weighted SD as $sdw=∑i=1Nwixi−x¯w2N′−1∑i=1NwiN′$⁠. We conducted meta-analysis with a random-effects model using Review Manager for outcomes with at least two included studies (RevMan 5.4; The Nordic Cochrane Centre, Cochrane, Copenhagen, Denmark). We applied the inverse variance method to all continuous outcomes and weighted outcomes by the number of eyes. We applied the Mantel-Haenszel method to discrete outcomes. Regarding between-study heterogeneity, we considered I² values greater than 50% to be associated with substantial statistical heterogeneity and greater than 75% to be associated with considerable statistical heterogeneity [12]. Methodology outlined per Beck et al. [13] was used to calculate Snellen equivalents for logMAR values. We converted BCVA values from early treatment diabetic retinopathy study (ETDRS) letters to logMAR using methodology outlined in Khoshnood et al. [14]. A change in 5 ETDRS letters was equivalent to a change in logMAR of 0.1 or 1 Snellen line [13]. We performed a leave-one-out sensitivity analysis to assess the robustness of our findings. A p value <0.05 was considered statistically significant and all p values were two-sided.

A total of 7,686 studies underwent title and abstract screening, 395 underwent full-text screening and 19 studies reporting on 1,291 eyes at baseline were included in our meta-analysis (Fig. 1) [3, 15‒32]. Six included studies were RCTs [17, 23, 29‒32], and 13 were observational studies [3, 15, 16, 18‒22, 24‒28]. All patients also underwent ERM removal. In one study, patients in the ILM peel group received whole-piece ILM peeling or maculorrhexis ILM peeling [24]. In one study, patients in the non-ILM peel group received fovea-sparing ILM peel [31]. Seven studies used ICG to stain the ILM [3, 15, 16, 24‒27], two studies used ICG or TA [20, 28], one used ICG or trypan blue [22], two used trypan blue [30, 31], one used trypan blue or brilliant blue [17], five used brilliant blue [18, 19, 23, 29, 32], and one used brilliant blue or TA [21]. The mean age of patients across treatment groups ranged from 62 to 77 years old. The proportion of males ranged from 5% to 90%. The mean baseline Snellen BCVA was 20/63 in the ILM peel group and 20/63 in the non-ILM peel group. The mean initial RT in the ILM peel group was 467.53 ± 95.86 μm and in the non-ILM peel group was 462.93 ± 96.82 μm. Five studies reported on central foveal thickness [3, 20, 23, 24, 29], five on CMT [16, 18, 19, 25, 32], and five on central retinal thickness [17, 26‒28, 31]. A complete summary of baseline characteristics and demographics can be found in Table 1.

Our risk of bias assessment for RCTs and observational studies are summarized in online supplementary eTables 2 and 3, respectively. Using the ROB2 tool, our risk of bias assessment of six included RCTs found that 35 domains (83.3%) had a low risk of bias and seven domains (16.7%) had a moderate risk of bias. The following represents the proportion of RCTs with low risks of bias across domains of the ROB2 tool: missing outcome data (100%), conflicts of interest (100%), industry sponsorship (100%), randomization process (83.3%), deviation from intended interventions (66.7%), measurement of the outcome (66.7%), and selection of reported results (66.7%). Using the ROBINS-I tool, our risk of bias assessment of 13 observational studies found that 67 (73.6%) domains had a low risk of bias and 24 (26.4%) domains had a moderate risk of bias. The proportion of observational studies with low risk of bias across domains of the ROBINS-I tool was as follows: classification of interventions (100%), measurement of outcomes (100%), selection of the reported results (100%), missing data (92.3%), selection of participants into study (92.3%), deviations from intended interventions (30.8%), and confounding (0%).

Out of 18 outcomes in the primary analysis, one (5.6%) was associated with a low certainty of evidence, 16 with a moderate certainty of evidence (88.9%), and one with a high certainty of evidence (5.6%). Our certainty of evidence for each outcome in the primary analysis and subgroup analyses is summarized in online supplementary eTable 4.

Nineteen studies reported one or more outcomes comparing eyes receiving PPV with and without ILM peel [3, 15‒32]. Across 17 studies, eyes receiving PPV with ILM peel (0.30 ± 0.31 logMAR, ∼20/40 Snellen, n = 569) and without ILM peel (0.30 ± 0.31 logMAR, ∼20/40 Snellen, n = 578) achieved a similar BCVA at last study observation (WMD = −0.01 logMAR, 95% CI = [−0.04, 0.03], p = 0.68, Fig. 2a). BCVA was similar between ILM peel and non-ILM peel groups at 3 months (p = 0.11, Fig. 2b), 6 months (p = 0.53, Fig. 2c), and 12 months (p = 0.30, Fig. 2d).

Across 11 studies, eyes receiving PPV with ILM peel (−0.30 ± 0.31 logMAR, n = 275) and without ILM peel (−0.25 ± 0.26 logMAR, n = 252) had a similar change in BCVA from baseline to last study observation (WMD = 0.01 logMAR, 95% CI = [−0.01, 0.02], p = 0.36, Fig. 3a). Change in BCVA from baseline was similar between ILM peel and non-ILM peel groups at 3 months (p = 0.51, Fig. 3b), 6 months (p = 0.10, Fig. 3c), and 12 months (p = 0.71, Fig. 3d).

Across 13 studies, eyes receiving PPV with ILM peel (340.92 ± 68.07 μm, n = 380) had a significantly thicker RT than eyes receiving PPV without ILM peel (335.78 ± 71.18 μm, n = 423) at last study observation (WMD = 11.97 μm, 95% CI = [1.64, 22.30], p = 0.02, Fig. 4a). RT was significantly thicker in the ILM peel group than the non-ILM peel group at 3 months (WMD = 20.73 μm, 95% CI = [1.45, 40.02], p = 0.04, Fig. 4b) and 6 months (WMD = 17.67 μm, 95% CI = [14.38, 20.96], p < 0.00001, Fig. 4c), but not 12 months (p = 0.51, Fig. 4d).

Across eight studies, eyes receiving PPV with ILM peel (−115.76 ± 98.21 μm, n = 231) and without ILM peel (−112.04 ± 81.58 μm, n = 209) achieved a similar change in RT from baseline to last study observation (WMD = 2.30 μm, 95% CI = [−10.73, 15.33], p = 0.73, Fig. 5a). Change in RT from baseline was similar between ILM peel and non-ILM peel groups at 3 months (p = 0.20, Fig. 5b), 12 months (p = 0.10, Fig. 5c). However, the ILM peel group (−123.75 ± 94.57 μm, n = 130) had a significantly smaller change in RT than the non-ILM peel group (−126.22 ± 73.59 μm, n = 108) at 6 months (WMD = 10.48 μm, 95% CI = [6.41, 14.54], p < 0.00001, Fig. 5d).

Across 13 studies, eyes receiving PPV with ILM peel had a significantly lower incidence of ERM recurrence (2.7% vs. 15.7%, RR = 0.26, 95% CI = [0.13, 0.51], p < 0.0001, Fig. 6a). The incidence of additional surgery for ERM was significantly lower in the ILM peel group compared to the non-ILM peel group (0.3% vs. 4.1%, RR = 0.17, 95% CI = [0.04, 0.74], p = 0.02, Fig. 6b).

In this subgroup, eight studies reported one or more outcomes comparing eyes receiving PPV with and without ILM peel, where patients had not undergone phacoemulsification simultaneously at the time of PPV or within 1 month of PPV [3, 17‒23]. All outcomes were consistent with the primary analysis, except for the outcomes described herein. Across two studies, RT was significantly thicker in the ILM peel group at 12 months (WMD = 23.06 μm, 95% CI = [3.12, 43.01], p = 0.02). Across two studies, the ILM peel group achieved a superior change in BCVA from baseline at 3 months (WMD = −0.08 logMAR, 95% CI = [−0.14, −0.01], p = 0.03). Change in BCVA from baseline to 6 months and 12 months, change in RT from baseline to 6 months and 12 months, and the incidence of additional surgery for ERM recurrence could not be analyzed due to a paucity of data.

In this subgroup, 11 studies reported on one or more outcomes comparing eyes receiving PPV with and without ILM peel, where studies included patients receiving simultaneous phacoemulsification [15, 16, 24‒32]. All outcomes were consistent with the primary analysis, except for the outcomes described herein. Across eight studies, RT was similar between the ILM peel and non-ILM peel groups at last study observation (p = 0.25). Across two studies, RT was similar between the ILM peel and non-ILM peel groups at 3 months (p = 0.36). Across three studies, the incidence of additional surgery for ERM recurrence was similar between the ILM peel and non-ILM peel groups (p = 0.12). Change in BCVA and RT from baseline to 3 months could not be analyzed due to a paucity of data.

In this subgroup, six RCTs reported on one or more outcomes comparing eyes receiving PPV with and without ILM peel [17, 23, 29‒32]. All outcomes were consistent with the primary analysis, except for the outcomes described herein. Across four RCTs, the non-ILM peel group achieved a superior change in BCVA from baseline to last study observation (WMD = 0.02 logMAR, 95% CI = [0.00, 0.04], p = 0.03). Across two RCTs, the non-ILM peel group achieved a superior change in BCVA from baseline to 12 months (WMD = 0.02 logMAR, 95% CI = [0.00, 0.04], p = 0.04). BCVA and RT at 3 months, change in BCVA and RT at 6 months, the incidence of repeat additional surgery for ERM recurrence could not be analyzed due to a paucity of data.

In this subgroup, 13 observational studies reported on one or more outcomes comparing eyes receiving PPV with and without ILM peel [3, 15, 16, 18‒22, 24‒28]. All outcomes were consistent with the primary analysis, except for the outcomes described herein. Across ten studies, RT at last study observation was similar between ILM peel and non-ILM peel groups (p = 0.09). Across four studies, the ILM peel group achieved a superior BCVA at 6 months (WMD = −0.05 logMAR, 95% CI = [−0.06, −0.04], p < 0.00001). Across four studies, RT at 3 months was similar across ILM peel and non-ILM peel groups (p = 0.17). Across five studies, the non-ILM peel group achieved a superior change in RT from baseline to last study observation (WMD = 10.14, 95% CI = [6.08, 14.20], p < 0.00001). Change in BCVA and RT at 3 months and 12 months could not be analyzed due to a paucity of data.

All leave-one-out sensitivity analyses were consistent with the primary analysis, except for in the following outcomes. RT at last study observation became similar between comparators when Lee et al. [27] (p = 0.09) and Ripandelli et al. [23] (p = 0.08) were excluded. The incidence of additional surgery for ERM recurrence became similar between comparators when Shimada et al. [21] (p = 0.12) were excluded.

The following outcomes from the primary analysis were associated with substantial heterogeneity: BCVA at last study observation (I² = 66%), BCVA at 6 months (I² = 55%), change in BCVA at 6 months (I² = 97%), and change in RT at last study observation (I² = 54%).

Our systematic review and meta-analysis investigated the safety and effectiveness of PPV with and without ILM peeling for the treatment of ERM. We found that the addition of ILM peeling did not improve visual acuity in ERM eyes. Patients who did not receive ILM peeling had a statistically significant reduction in RT at last study observation in RCTs, although this did not hold true in observational studies, and the magnitude of the difference between groups was small and it is unclear if this difference is clinically meaningful. Nevertheless, the ILM provides a scaffold for transdifferentiated Müller cell and myofibroblast proliferation; thus, the risk of recurrent ERM is reduced when the ILM is removed [2, 21]. Hence, our study found that ILM peeling was associated with a lower incidence of ERM recurrence and need for repeat surgery, which is consistent with the analysis of Fang et al. [33].

A previous meta-analysis by Far et al. [34] found that PPV with and without ILM peel for ERM achieved similar final visual acuities in RCTs. In accordance with Far et al. [34], our study found that PPV with and without ILM peel achieved similar visual acuities at last study observation, irrespective of whether studies administered phacoemulsification simultaneously with PPV. Expanding upon this previous work, our meta-analysis also analyzed the change in BCVA from baseline to last study observation in RCTs, finding that eyes treated without ILM peeling had superior improvements in BCVA compared to those receiving ILM peeling. The significance of this outcome was strongly driven by the low SD reported in Russo et al. [31] and it would not have been significant if this study was excluded from our subgroup analysis. Furthermore, when analyzing RCTs and observational studies together, we found that ILM peel and non-ILM peel groups had similar changes in BCVA from baseline.

Far et al. [34] found that ILM peeling was associated with an increased CMT at 12 months, which may have been mediated by ultrastructural damage to the inner retina secondary to ILM peeling. In contrast, our study found that ILM peeling resulted in a statistically significantly thicker RT earlier in the postoperative period at 3 and 6 months, although it is unclear if the difference between groups was clinically meaningful. It is postulated that swelling of the retinal nerve fiber layer during the early postoperative period following ILM peeling may contribute to an increased RT [35]. This may occur because removal of the ILM effectively removes a barrier for water transport between the inner retina and vitreous humor, facilitating fluid flow into the inner retina [35‒37]. Another meta-analysis in this setting found that eyes treated without ILM peel had greater reductions in CMT before 12 months, although this difference became nonsignificant in the long term at 12 months and beyond [38]. These findings align with our meta-analysis, as we found that differences in RT from baseline became similar between the ILM peel and non-ILM peel groups at 12 months and last study observation. Our study further added that PPV without ILM peel had a greater reduction in RT at 6 months, although comparators had similar changes in RT at 3 months.

Our meta-analysis also found that the non-ILM peel group overall achieved a lower RT at last study observation than the ILM peel group, supporting the findings of Far et al. [34]. Nonetheless, ten out of 13 studies reporting on RT at last study observation included patients with a slightly greater RT at baseline in the ILM peel group relative to the non-ILM peel group. The significance of this outcome was also driven by the findings of Lee et al. [27], and it would not have been significant if this study was excluded from our primary analysis. In our subgroup of ten observational studies, RT at last study observation became similar between ILM peel and non-ILM peel groups. A meta-analysis by Azuma et al. [39] identified no RCTs and found that postoperative CMT was similar between ILM peel and non-ILM peel groups across five studies.

A paucity of RCT evidence and low sample sizes across outcomes limited the strength of our study, resulting in very low to moderate certainties of evidence. As well, the significance of some outcomes in our meta-analysis changed after a leave-one-out sensitivity analysis. The substantial heterogeneity associated with a few outcomes limits the generalizability of our conclusions. The use of adjuvants for ILM staining differed across studies and may have also influenced outcomes. For instance, it is plausible that ICG toxicity may have affected visual acuity [40], however corroborating data for this was not directly provided by the included studies. Moreover, trypan blue and TA are not considered ILM-specific [2], which may lead to incomplete ILM peeling. Inconsistent reporting of safety and effectiveness outcomes limited the number and robustness of comparisons. Different studies reported on BCVA or RT at last study observation and change in BCVA or RT from baseline, resulting in inconsistent findings. Metamorphopsia, which tended to occur in ERM eyes undergoing PPV with or without ILM peeling, could not be analyzed due to the reporting of different metrics across studies [30, 32]. Some eyes that did not receive simultaneous phacoemulsification were included in the simultaneous phacoemulsification subgroup analysis, as many included studies reporting on eyes receiving simultaneous phacoemulsification did not provide data stratified by this criterion [16, 24‒30, 32]. Studies such as Russo et al. [31] and Lee et al. [27] reported small SDs and were heavily weighted in the random-effects meta-analysis, likely driving the significance of several outcomes. Our meta-analysis found that PPV without ILM peeling was associated with a lower RT than PPV with ILM peeling but a higher incidence of ERM recurrence. Although recurrent ERMs may be indicative of an increased RT, it is possible that the incidence of ERM recurrence (2.7% in the ILM peel group and 15.7% in the non-ILM peel group) was too low in the non-ILM peel group to meaningfully increase the group’s mean RT and future research is needed to better understand the relationship between RT outcomes and ILM peeling. Results of this study should be interpreted only at the level of the cohort and not at the level of individual patients. Conclusions must be regarded as hypothesis-generating.

In conclusion, our meta-analysis found that PPV with and without ILM peel offer similar improvements to visual acuity in patients with ERM. PPV without ILM peel achieved a reduced RT at last study observation. Nevertheless, patients treated with PPV and ILM peel had a lower recurrence of ERM and lower rates of additional surgery. Given the moderate certainty of evidence associated with many of our outcomes, we recommend that future RCTs evaluate the safety and effectiveness of PPV with and without ILM peel in diverse patient populations to allow for more nuanced treatment decision-making.

An ethics statement is not applicable because this study is exclusively based on published literature. The study is in accordance with the World Medical Association Declaration of Helsinki.

Andrew Mihalache, Ryan S. Huang, Haleema Ahmed, and Nikhil S. Patil: none. Marko M. Popovic: financial support (to institution) – PSI Foundation and Fighting Blindness Canada. Rajeev H. Muni: consultant – Alcon, Apellis, AbbVie, Bayer, Bausch Health, and Roche; and financial support (to institution) – Alcon, AbbVie, Bayer, Novartis, and Roche. Peter J. Kertes: honoraria – Novartis, Bayer, Roche, Boehringer Ingelheim, RegenxBio, and Apellis; advisory board – Novartis, Bayer, Roche, Apellis, Novelty Nobility, Viatris, and Biogen; and financial support (to institution) – Roche, Novartis, Bayer, and RegenxBio.

No specific grants were received from any funding agency for this research.

Andrew Mihalache, Ryan S. Huang, Haleema Ahmed, Nikhil S. Patil, Marko M. Popovic, Peter J. Kertes, and Rajeev H. Muni have satisfied the four criteria of authorship detailed by the International Committee of Medical Journal Editors (ICMJE) recommendations. Andrew Mihalache, Ryan S. Huang, Haleema Ahmed, Nikhil S. Patil, Marko M. Popovic, Peter J. Kertes, and Rajeev H. Muni contributed substantially to one or more of the following aspects of the manuscript: conception of the study, data acquisition, statistical analysis, drafting, revision for intellectual content, and final approval.

All generated data are included in this paper and its supplementary material files. Further inquiries are to be directed to the corresponding author.