Retinal Detachment After Anti-VEGF Injections: How Rare Is the Risk? A Critical Meta-Analysis

Question: How frequently does retinal detachment (RD), particularly rhegmatogenous retinal detachment (RRD), occur following intravitreal anti–vascular endothelial growth factor (anti-VEGF) injection?

Design: PRISMA systematic review and random-effects single-arm meta-analysis of proportions, prospectively registered as PROSPERO CRD420251175527. MEDLINE, Embase and Cochrane CENTRAL were searched through January 18, 2026. Twelve observational studies contributed 834,814 injections and 119 reported RD events. Nine studies with clearly characterized RRD contributed 787,849 injections and 92 RRD events to the primary analysis.

Key distinction: This estimates an event proportion among injections, not a causal excess risk attributable to injection versus an unexposed comparator.

The pooled RRD proportion was 0.012% per injection (95% CI 0.009%–0.015%; I² 14.4%), or approximately one event per 8,675 injections. The authors used a binomial generalized linear mixed model with logit transformation, allowing zero-event cohorts to contribute without a continuity correction. Leave-one-out sensitivity analyses yielded estimates from 0.0105% to 0.0128%, suggesting that no single study dominated the pooled estimate.

These are population-level pooled estimates from heterogeneous observational cohorts; they are not an individualized risk prediction, nor an incidence rate standardized by follow-up time.

A sensitivity analysis restricted to five studies with an explicitly defined postprocedural window of 90 days or less estimated RRD at 0.013% per injection (95% CI 0.010%–0.016%; I² 36.8%), or approximately 1 in 7,692 injections.

Three studies with eye-level denominators yielded 0.08% per eye (95% CI 0.02%–0.31%; I² 66.7%), approximately 1 in 1,250 treated eyes. Those eyes received a pooled mean of 13.8 injections. This per-eye estimate captures a different denominator and exposure pattern from the per-injection estimate; it must not be treated as the lifetime risk for all patients receiving years of therapy.

The authors deliberately excluded tractional retinal detachments (TRD) from the primary RRD meta-analysis. In one cohort of eyes with severe proliferative diabetic retinopathy (PDR) treated with preoperative bevacizumab, reported TRD was 3.6% (95% CI 2.3%–5.2%), approximately one in 28 injections. This is a single disease-specific cohort estimate, not the general anti-VEGF injection risk.

Fibrovascular contraction after anti-VEGF therapy in advanced PDR provides a different proposed mechanism from a possible mechanically associated retinal break following an intravitreal injection. The pooled RRD estimate and severe-PDR TRD proportion should never be combined or compared as if they arose from equivalent populations.

Exploratory comparisons did not detect statistically significant subgroup differences by injection volume (P = .520), single- versus multicenter setting (P = .258), nAMD-only versus mixed indications (P = .061), provider type (P = .259), or risk-of-bias strata (P = .356). These subgroup analyses involved few studies and were underpowered; lack of statistical significance is not evidence of identical safety across indications, providers or techniques.

Crucial potential risk modifiers—axial length, lens status, vitreoretinal interface/PVD status, needle gauge, quadrant and injection technique—were incompletely documented, preventing reliable risk stratification.

The paper discusses one cohort in which cumulative RRD probability rose from 0.15% after 10 injections to 0.54% after 50 injections during long-term follow-up, without evidence that risk per individual injection increased with successive exposures (P = .95). Another cohort did not find an association with number of injections, but lifetime exposure was incompletely captured.

These observations are compatible with low risk per procedure and greater cumulative opportunity for an event as injections accumulate. They do not prove independence between injections or yield a universal lifetime risk curve. Per-eye and per-injection probabilities should be presented separately in patient counseling.

In some included case series, retinal tears occurred near the injection quadrant, and events were recognized from days to weeks after injection. The authors interpret these patterns as consistent with a possible focal mechanical mechanism, but observational timing and location cannot establish causation. Underlying retinal pathology may also contribute.

Some studies reported a substantial proportion of macula-off detachments at diagnosis. Scheduled injection follow-up therefore does not replace clear instructions to seek urgent ophthalmic assessment for new flashes, sudden floaters, a curtain or shadow in the field of vision, or acute visual deterioration.

• Informed consent: Describe RRD as rare after anti-VEGF injection, with a pooled estimate near 0.012% per injection, while acknowledging that patient-specific risk is not well quantified.
• Long-term therapy: Explain that per-eye cumulative exposure is not represented by a single per-injection percentage; the eye-level estimate is less precise.
• Severe PDR: Discuss TRD separately when substantial fibrovascular proliferation is present, particularly in the preoperative anti-VEGF context.
• Safety-net advice: Give explicit warning symptoms and a route to immediate evaluation; the review does not establish an optimal routine screening interval for asymptomatic patients.
• Technique: Continue evidence-based injection protocols, but do not claim this meta-analysis identifies a safest quadrant, needle gauge or provider category.

• Low to very low certainty: The underlying evidence was observational; seven studies were retrospective cohorts, four were retrospective case series and one was a prospective cohort. GRADE certainty was low or very low.
• Ascertainment bias: RD treated elsewhere or diagnosed beyond a study’s follow-up window may be missed. Only five studies reported time to RD detection, and follow-up duration was often incomplete or inconsistent.
• Denominator mismatch: A pooled per-injection proportion is not a person-time incidence rate, nor the probability that a particular eye will detach after an entire treatment course. Repeated injections within eyes introduce dependence and heterogeneous exposure.
• Clinical heterogeneity despite low I²: Indications, anti-VEGF agents, provider training and injection protocols differed; many potentially important covariates were not recorded in the source populations.
• No causal comparator: The study cannot distinguish injection-attributable detachments from background risk due to pre-existing ocular disease or other factors.
• Rare events and sparse subgroups: Confidence intervals and subgroup comparisons have limited power; the per-eye estimate has substantial heterogeneity (I² 66.7%).
• Incomplete retinal-break reporting: Studies did not quantify retinal tears treated before progressing to detachment, so RD events alone may not capture all relevant retinal complications.

RRD following intravitreal anti-VEGF injection is uncommon in published observational series, with a pooled event proportion of approximately one in 8,675 injections. The more patient-centered eye-level estimate is less precise and cannot be generalized to indefinite treatment exposure. The markedly different TRD estimate from one severe-PDR cohort should be considered separately. The most defensible clinical use of these findings is calibrated informed consent, disease-specific counseling and prompt symptom-triggered evaluation—not overconfident individual risk prediction or claims of equivalence among injection techniques.