Neuro-Ophthalmic Toxicity from Immune Checkpoint Inhibitors

Key Takeaways

• Food and Drug Administration Adverse Event Reporting System reports of ICI-associated neuro-ophthalmic irAEs were more often submitted for males and for adults aged 65 years or older.
• Pembrolizumab and avelumab were the only ICIs to meet HLGT-level criteria for a positive neuro-ophthalmic safety signal.
• Six PT-level neuro-ophthalmic signals were identified, and eyelid ptosis dominated the reported event profile.
• Reported neuro-ophthalmic events tended to occur early after ICI exposure, and hospitalization was the most frequently recorded outcome among serious reports.

Ptosis, diplopia, ophthalmoplegia, and related neuro-ophthalmic toxicities can emerge during immune checkpoint blockade at the overlap of ophthalmology, neurology, and oncology, where they may resemble other ocular motility or neuromuscular syndromes. That diagnostic overlap can complicate recognition when visual symptoms develop during active cancer treatment and several neurologic explanations remain plausible. Postmarketing signal detection can help map which phenotypes are being reported, when they appear, and how often they are linked to severe outcomes.

In Li et al. FAERS analysis of neuro-ophthalmic toxicity with immune checkpoint inhibitors, investigators performed a retrospective pharmacovigilance analysis using the Food and Drug Administration Adverse Event Reporting System (FAERS) from Q1 2011 through Q4 2025 and limited the dataset to 521 reports in which an immune checkpoint inhibitor (ICI) was coded as the primary suspect drug after Food and Drug Administration-recommended deduplication.

Neuro-ophthalmic cases were defined through Medical Dictionary for Regulatory Activities (MedDRA) version 26.1 High-Level Group Term (HLGT) ocular neuromuscular disorders with Preferred Term (PT) validation. Signal detection used the Reporting Odds Ratio (ROR) and Bayesian information component (IC), with positive signals requiring ROR025 greater than 1, IC025 greater than 0, and at least 3 reports against all other drugs and events in FAERS as the comparator. This approach identifies disproportional reporting rather than bedside causation.

At the HLGT level, positive neuro-ophthalmic signals for pembrolizumab and avelumab were the only ones to meet study thresholds: pembrolizumab had ROR 1.45 (1.3–1.62) and IC 0.53 (0.37), and avelumab had ROR 2.12 (1.23–3.65) and IC 1.08 (0.2). Only pembrolizumab and avelumab met the prespecified HLGT signal thresholds, although the avelumab estimate was based on very small numbers and wide confidence intervals and may be unstable.

PT-level analysis identified six positive signals, and eyelid ptosis accounted for 426 of 614 PT-level reports (69.4%). Pembrolizumab and nivolumab each showed a broad PT signal pattern, and durvalumab had the strongest ocular myasthenia signal. The reported phenotype profile therefore centered on ptosis and related ocular neuromuscular complaints.

Among 204 of 521 reports with usable onset data, median onset ranged from 28 to 38 days, with most events clustering around 5 weeks after treatment initiation. Acute presentations were reported with pembrolizumab, nivolumab, and atezolizumab, whereas avelumab and durvalumab showed a later pattern; pembrolizumab and nivolumab also appeared in longer-latency reports.

Hospitalization was the most frequent reported outcome, and serious outcomes occurred in 28.8% of reports. Pembrolizumab and nivolumab had the highest reported fatal-case proportions in this dataset.

As a spontaneous-report database, FAERS cannot establish causality, incidence, or comparative clinical risk, and it does not provide exposure denominators, causality verification, or key confounders such as tumor type, treatment duration, line of therapy, or combination regimens. Timing conclusions rest on a minority of reports, and some agent-level and PT-level estimates, especially for avelumab and cemiplimab, were unstable because they were based on very small samples and wide confidence intervals. No submitted FAERS reports documented recovery or resolution, which limits inference about downstream course. The findings are therefore best read as a postmarketing description of what has been reported, not as a measure of true event rates.

The authors concluded that neuro-ophthalmic reports were concentrated in older adults and men, often emerged early after immune checkpoint inhibitor exposure, and most clearly signaled at the HLGT level for pembrolizumab and avelumab, with ptosis as the dominant manifestation. They framed these findings as hypothesis-generating postmarketing signals that require external validation.

Clinician Questions

What counted as a positive neuro-ophthalmic safety signal in the FAERS ICI analysis?

A positive signal required all 3 prespecified thresholds: ROR025 greater than 1, IC025 greater than 0, and at least 3 reports. In FAERS, that framework identifies disproportional reporting relative to all other drugs and events rather than proving causality or incidence.

Why was atezolizumab left out of the PT-level neuro-ophthalmic signal analysis?

Atezolizumab had fewer than 3 neuro-ophthalmic Preferred Term reports meeting positive-signal criteria, so it was excluded from the PT-level analysis. That exclusion also illustrates how sparse reporting can leave some drug-event estimates unstable.

How complete were onset data for ICI-related neuro-ophthalmic irAEs in FAERS?

Only a minority of FAERS reports had usable onset dates, so the median onset range and the acute-versus-delayed timing pattern were derived from a subset rather than the full database. That limitation makes the timing signal clinically suggestive rather than definitive.

Were recovery or resolution outcomes documented in FAERS reports of ICI-related neuro-ophthalmic irAEs?

No. Recovery or resolution was not documented in the submitted FAERS reports, so the database could not clarify downstream course or prognosis.