Association of Individual-Level Factors With Visual Outcomes in Optic Neuritis

This secondary analysis of a randomized clinical trial develops models of individual-level visual outcomes for patients with optic neuritis.


Introduction
The Optic Neuritis Treatment Trial (ONTT) was a landmark randomized clinical trial showing that treatment with corticosteroids does not improve long-term visual outcomes for patients with acute demyelinating typical optic neuritis but may hasten visual recovery in the first 2 weeks. [1][2][3] Given the lack of a robust treatment effect with corticosteroids, the ONTT Study Group and clinical practice guidelines called for shared decision-making between patients and physicians when deciding whether to use corticosteroids for acute neuritis. 4 Since the publication of the ONTT findings, the diagnosis of typical optic neuritis remains unchanged, and no additional short-term treatments have emerged. Therefore, the ONTT provides the only actionable data to date for counseling patients on the benefits of corticosteroids.
Despite the lack of corticosteroid effect on long-term outcomes, more than 90% of neurologists report treating patients with optic neuritis with corticosteroids. 5 Patient preference was the least cited reason for treatment, suggesting shared decision-making is not occurring. 5,6 One barrier to the application of shared decision-making is that clinicians do not have information that considers individual characteristics, such as severity of the visual deficit, and how those individual characteristics might influence long-term recovery or likelihood to experience a hastened recovery. 5 Personalizing clinical trial information is a strategy used to address this barrier and ultimately promote informed, shared decision-making. 7 Currently, we know of no studies that allow clinicians to provide personalized information regarding visual recovery and short-term benefit with corticosteroids to patients with optic neuritis.
The purpose of our study was to perform a multivariable, risk-stratified analysis of ONTT data to assess visual function over time and at 1 year at the individual level. We hypothesized that several baseline factors would be associated with vision outcomes over time. Ultimately, this information could support shared decision-making.

Standard Protocol Approvals, Registrations, and Patient Consent
Because ONTT data are deidentified and publicly available, 8 our study was determined to have "not regulated" status by the University of Michigan's institutional review board in accordance with federal regulations regarding human subjects research. A waiver of consent was not required. The reporting of this study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. 9

Study Population
We conducted a secondary analysis of data from the ONTT (data downloaded on October 15, 2018), which enrolled patients from July 1988 to June 1991 and then continued to follow up patients for 15 years. Analyses were performed from January 24, 2019, to February 20, 2020. The ONTT is the most comprehensive data source for optic neuritis because of the large number of patients, wide range of clinically important patient variables collected, systematic measurement of baseline variables, and low attrition rates.
The ONTT enrolled 457 people at 14 academic eye centers and 1 large community eye center.
Participants were aged 18 to 46 years with incident acute unilateral optic neuritis within 8 days of onset of vision loss as determined by the neuro-ophthalmologist site investigator using standard criteria. 1,6 Study visits after enrollment occurred at 4, 15, and 30 days and 7, 13, 19, 26, and 52 weeks.
Yearly visits occurred thereafter. 6 neuritis episodes in the fellow eye, days of vision loss, pain, optic disc swelling, viral illness within 1 month, and baseline VA (for the VA model only) or baseline contrast sensitivity (CS; for the CS model only). Given the limited magnetic resonance imaging data, it was not included as an independent variable.
Visual acuity was measured in the ONTT using retroilluminated Early Treatment of Diabetic Retinopathy Study (ETDRS) charts and converted to the logarithm of the minimal angle of resolution (logMAR) for analysis purposes. A standard logMAR equivalent for patients with VA of count fingers (CF), hand motion, light perception, and no light perception does not exist. 11,12 The original ONTT analysis assigned a logMAR VA of 1.7 for all participants with VA of CF or less. For analyses requiring numeric values, we used 1.85 for CF, 2.30 for hand motion, 2.75 for light perception, and 3.20 for no light perception. These measurements are similar to previously used values and permitted easy display of the data. 11 Conclusions from models were insensitive to our choices. For reporting results, all logMAR values were converted back to Snellen equivalents. Contrast sensitivity was measured in the ONTT using Pelli-Robson charts, which use letters that subtend 2.8°at 1 m. Letters are arranged in groups of 3 with 16 steps of decreasing contrast of 0.15 log units, with each step relative to the chart background (range, 0-16 steps).

Outcome Variables
Our primary outcome was VA at 1 year. Secondary outcomes were CS at 1 year and VA and CS at days 15 and 30.

Statistical Analysis and Prediction Models
Demographic and baseline clinical characteristics of participants who did not complete the 1-year visit were compared with those who did using 2-tailed t tests for continuous data and Fisher exact test for categorical data; 2-sided P < .05 indicated significance. Multiple linear regression models were built to estimate VA and CS at 15 and 30 days and 1 year using a prespecified plan to evaluate model assumptions. The models were built using all available ONTT data at each point. Model discrimination, which is the variability of the outcome measure explained by the independent variables, was estimated using adjusted R 2 values. Calibration, which is an assessment of how closely estimated values represent true values, was assessed using numeric and graphical analysis.
For the VA models, first we used ordinary least squares regression; however, regression diagnostics revealed multiple violations of ordinary least squares regression assumptions, including residuals with heteroscedastic variance; nonlinearity of the association between the independent and outcome variables; and nonnormal, skewed residuals. Robust regression (M-estimation) and generalized additive models were considered, but both still violated assumptions when using all observations. After exploring our data, we determined that patients with very low baseline acuity were driving model assumption violations, and so we stratified the data (VA better than CF vs CF or worse) for analysis. Analysis within these subsets did not demonstrate the noted violations. Therefore, separate subset linear regression models were used to estimate VA outcome in those participants with a baseline VA better than CF and the estimated VA outcome in those participants with a baseline VA of CF or worse. Likewise, separate subset linear regression models were used to estimate CS outcome in those participants with a baseline CS better than 0 and those with baseline CS of 0. We performed sensitivity analyses using dichotomous outcomes of 20/20 or better and 20/40 or better for VA and CS of 12 or better. Results were similar to the analyses presented.
Heterogeneity of treatment effect was investigated for both the VA and CS models by including interactions between baseline vision and treatment group. The 95% predictions intervals for individuals were made based on quantile regressions of the response variable on baseline VA or CS.

JAMA Network Open | Neurology
Association of Individual-Level Factors With Visual Outcomes in Optic Neuritis  Our study builds on prior work of predicted long-term visual outcomes and treatment benefit. Our study can be used to inform treatment decisions for patients with typical optic neuritis.

JAMA Network Open | Neurology
To apply our results to estimate the 15-day and 1-year VA in patients at the time of baseline evaluations, clinicians can make the specific calculation using the linear regression formula in Tables 2   and 3 or, more simply, use our Table 4 Table 4 could also draw a clinician's attention to the fact that a patient may not have typical optic neuritis if they are not recovering in the expected manner during the first month.
Because the early benefit of corticosteroids was small when present at all, patients and clinicians should weigh this information against the potential harms of corticosteroids. In the ONTT, approximately 50% of patients experienced at least mild adverse effects, such as insomnia. 15 Other inconveniences of corticosteroids were not captured, including hospitalizations, travel to infusion centers, or difficulty taking more than 20 tablets per day of prednisone if a high-dose oral regimen was used. 16 Among patients using short-term, low-dose corticosteroids (median prednisone

Limitations
Our study has important limitations. First, the ONTT consisted primarily of women who were white with typical demyelinating optic neuritis. Therefore, our findings cannot be generalized to patients with atypical optic neuritis, including neuromyelitis optica and myelin oligodendrocyte glycoproteinassociated optic neuritis or those who would not have met the ONTT inclusion criteria. We also did not find an association between symptom duration and long-term visual function but cannot address whether hyperacute corticosteroid treatment before the onset of vision loss would have aborted the attack. 18 Second, the ONTT data are more than 25 years old. Although little evidence suggests that the diagnosis or natural history of optic neuritis has changed substantially over time, external validation in a more modern cohort is needed. Third, few patients with poor visual function both at baseline and at 1 year were included. Further work should be undertaken to understand why some people with poor visual function do not substantially recover and to try to estimate who might be at risk. Third, we did not include optic neuritis recurrences as a variable in our models, because that information is unknown to the clinician at the time of initial presentation when corticosteroid treatment decisions are made. An unanticipated finding of the ONTT was that patients treated with oral corticosteroids were more likely to have an optic neuritis relapse. This finding is potentially important because those with a relapse were more likely to have a poor outcome. Fourth, MS is now diagnosed using more sensitive criteria (2017 McDonald criteria). 19 Therefore, the association between MS and optic neuritis visual outcomes should be evaluated in a study that can apply the 2017 McDonald criteria. Last, because the ONTT was underpowered to detect serious corticosteroid harms, we cannot use this data set to estimate risks of individual-level harms.

Conclusions
Shared decision-making is a central part of patient-centered health care and is particularly important when the benefits of an intervention have a marginal effect on outcomes and harms are an important concern. Personalizing the ONTT data to inform clinicians and patients about treatment benefits and long-term outcome has the potential to support both evidence-based care and shared decisionmaking. Further studies are needed to understand those patients with poor recovery and, importantly, to integrate information about corticosteroid benefits with estimates of corticosteroid harms.