Early vs Deferred Non–Messenger RNA COVID-19 Vaccination Among Chinese Patients With a History of Inactive Uveitis

Key Points Question Is there any difference in uveitis outcomes between patients with inactive disease given recommendations for early and deferred non–messenger RNA (mRNA) COVID-19 vaccination? Findings In this open-label, randomized clinical trial involving 543 patients, recommendation for early vaccination resulted in an increased incidence of self-reported worsening of symptomatic uveitis compared with deferred vaccination, but no differences were observed in disease and visual prognosis at 3 months. Meaning These findings suggest that recommendation for early non-mRNA COVID-19 vaccination in patients with inactive uveitis may lead to subjective uveitis symptoms but no adverse effects on disease and visual prognosis at 3 months.


eMethods. Instrumental Variable Analyses
The instrumental variable analysis was used to assess the actual effect of COVID-19 vaccination on the primary outcome. The randomization assignment (early vs deferred COVID-19 vaccination) was considered a valid instrumental variable based on the assumption that it affected the outcome only by modifying a person's probability of receiving COVID-19 vaccines and was otherwise unrelated to measured or unmeasured confounders for the outcome. The instrumental variable analysis tested the actual effect by first generating via linear regression that early vaccination recommendation increased COVID-19 vaccine exposure (β1). The assessment of effect of early vaccination recommendation on the primary outcome was obtained from the primary analysis (β2). Under the assumption that the entire effect of early vaccination recommendation on the primary outcome (β2) was mediated by its effect on increasing COVID-19 vaccine exposure (β1), the assessment of causal effect on the primary outcome per increase in COVID-19 vaccine exposure was obtained by β2/β1. For the purposes of our study, we defined the exposure in two separate analyses. In analysis 1, the exposure was defined as receipt of COVID-19 vaccine during the study period. In analysis 2, the exposure was defined as having been vaccinated in line with the early vaccination recommendation during the study period. The difference was that two vaccinated individuals, who were assigned to receive the deferred vaccination recommendation and achieved complete uveitis remission to be vaccinated, were not categorized into those who have been vaccinated in line with the early vaccination recommendation. In both analysis 1 and analysis 2, we defined the outcome as the time to uveitis symptomatic worsening, which was the primary outcome of this study. To assess the causal effect on the primary outcome per increase in each exposure, we used the two-stage estimation approach for analyzing time-to-event data to estimate the hazard ratio as well as its 95% CI as previously described. 1 The population included all participants who had undergone randomization.
Modified Intentionto-treat 511 The population included all participants who met eligibility criteria and had undergone randomization. A total of 32 participants who had been vaccinated before randomization were excluded from the intention-to-treat population.

Per protocol 351
The population included participants who indeed followed randomly assigned vaccination recommendation (early or deferred) with follow-up data and had no vaccination before randomization. A total of 160 individuals who did not adhere to randomly assigned recommendation were excluded from the modified intention-to-treat population.
Tele-follow-up completed population 506 The population included participants who completed month 3 follow-up by telephone call. A total of 37 participants who withdrew or were lost to telephone call follow-up before month 3 were excluded from the intentionto-treat population.
Month-3 in-person evaluable population 249 The population included participants who had completed month 3 in-person follow-up encounter. A total of 294 participants who did not attend month 3 in-person followup visit were excluded from the intention-to-treat population.

Multiple imputation
To examine the potential impact of under-reporting of events due to withdrawal or loss to follow-up, multiple imputations were used to predict missing values in 20 imputed datasets for primary outcome of those who withdrew or were lost to tele-follow up before month 3 in the intention-totreat population, based on the assumption that data were missing at random. Information about the recommendation allocation, completed follow-up encounters and all available values of baseline variables (age, sex, ethnic group, history of uveitis, etiology of uveitis, type of uveitis, best corrected visual acuity in the better seeing eye, number of flares in the past 12 months and medical history and comorbidities) was used for multiple imputations. The overall treatment effect hazard ratio was calculated by combining effects estimated from each imputed dataset in the Cox regression model with the Rubin's rule.

Multivariable
Cox model* Adjustment for known covariates may lead to increases in power. 3 Therefore, a Cox model as specified for the primary analysis in the intention-to-treat population was fit with further adjustment for all available baseline variables (age, sex, ethnic group, history of uveitis, etiology of uveitis, type of uveitis, best corrected visual acuity in the better seeing eye, number of flares in the past 12 months and medical history and comorbidities) in the sensitivity analysis.

Endpoint criteria modified*
In the sensitivity analysis, the endpoint criteria were modified and outcomes were re-adjudicated in the intention-to-treat population. Symptomatic uveitis worsening was re-defined if one of following newly onset symptoms occurred in at least one eye lasting for at least 2 days: eye redness, eye pain, decreased vision, or light sensitivity. Criterion of floaters was excluded to examine whether the results were sensitive to such a change in endpoint criteria.

Competing risks model*
Because a systemic event (eg. death or an adverse event of certain degree of severity) was likely to preclude the occurrence of primary endpoint event or greatly alter the chances to observe it, the competing-risks model (cumulative incidence function and Fine-Gray regression model) was used to account for the competing risk of systemic events in the intention-totreat population.
* These analyses were performed post-hoc. * These analyses aimed to assess the effect of vaccination (analysis 1) and the per protocol effect (analysis 2) on the primary outcome. In analysis 1, the exposure was defined as receipt of COVID-19 vaccine during the study period. In analysis 2, the exposure was defined as having been vaccinated in line with the early vaccination recommendation. The difference was that two vaccinated individuals, who were assigned to the deferred vaccination recommendation and achieved complete uveitis remission to be vaccinated, were not categorized into those who have been vaccinated in line with the early vaccination recommendation. Both outcomes were the time to uveitis symptomatic worsening. The randomization assignment (early versus deferred vaccination recommendation) was considered a valid instrumental variable based on the assumption that it affected the outcome only by influencing a person's probability of receiving COVID-19 vaccine and was otherwise unrelated to measured and unmeasured confounders for the outcome. The estimate of hazard ratio as well as its 95% CI was obtained with the use of the two-stage estimation approach for analyzing time-to-event data as previously described. 1 to 0.020) * The pseudopopulation was generated with the use of inverse probability weighting to account for unobserved data of participants who were not included in the month-3 in-person evaluable population. The probability of non-missing information at month 3 was predicted with the logistic regression model, where the response was the nonmissingness and the covariates included the occurrence of primary endpoint event during study period and all available values of baseline variables (age, sex, ethnic group, history of uveitis, etiology of uveitis, type of uveitis, best corrected visual acuity in the better seeing eye, number of flares in the past 12 months and medical history and comorbidities). The weight of each participant was given by the inverse of the predicted probability. Then, the analysis was performed only on the non-missing observations of the month-3 in-person evaluable population with a weighted model. Numbers of valid cases have been rounded. † Increases were relative to baseline condition. ‡ Changes in best corrected visual acuity were analyzed by eye with the generalized estimating equation to account for baseline values and the correlation between eyes of the same patient. Data are shown as least-squares means ± standard errors. Visual acuity data are expressed as scores for the log of the minimum angle of resolution (logMAR), with higher values indicating poorer vision.