Are adverse events higher among patients with acute optic neuritis prescribed glucocorticoids? A retrospective, longitudinal cohort study


  • Our large observational data set provided increased power and a robust array of potential confounding factors that we considered via high-dimensional propensity score adjustment.

  • Our study design assumes that the glucocorticoid was prescribed in response to the optic neuritis diagnosis given the proximity of the prescription and optic neuritis claim but could have been prescribed administered for another reason that was not fully documented in the data set.

  • The diagnosis of optic neuritis may be miscoded in the data set, although findings were unchanged in a sensitivity analysis using a more specific optic neuritis definition.

  • Because we used private US insurance claims, our findings may not be generalisable to those who are uninsured or part of a low-socioeconomic group.


Acute optic neuritis (ON) is an inflammatory optic nerve condition that causes vision loss in young adults and is routinely treated with high doses of glucocorticoids for 3–5 days, often followed by a taper.1 2 In the USA, ON is frequently idiopathic (30%) or associated with multiple sclerosis (MS) (57%).3 For these most common types of ON, either idiopathic or MS-associated, glucocorticoids provide no long-term benefits to visual functioning2 4 and the short-term benefits are marginal.4

Despite the routine use of glucocorticoids for ON, they pose a potential risk for harm. In the largest clinical trial to date of glucocorticoids for the treatment of ON, the Optic Neuritis Treatment Trial (ONTT), adverse events were frequent but mild among treated participants. Only two serious AEs were captured: acute pancreatitis and severe depression requiring hospitalisation.5 Both occurred in participants treated with glucocorticoids. Recent observational studies focused on providing real-world incidence rates of serious AEs among patients treated with short-term glucocorticoids for less than 2 weeks have found a 2-to-5 fold increase in serious AEs and events not previously captured from clinical trial data alone.6 7 This type of real-world information about the risk of AEs with glucocorticoids among patients with ON would complement clinical trial data so that patients and clinicians can fully consider treatment tradeoffs for the most common ON subtypes.

We used administrative data from a large, commercial insurer to assess the association between glucocorticoid prescriptions and possible AEs captured during healthcare encounters among patients with ON. Our large data set provided increased power and a robust array of potential confounding factors that we considered via high-dimensional propensity score adjustment. The results of our study can support decision making by providing clinicians and patients with real-world information about potential glucocorticoid-associated AEs.


Data source

We used Optum Clinformatics Data (Eden Prairie, MN), which contains detailed, deidentified claims for individuals insured by a large, commercial health insurance provider in the USA from 1 January 2004 to 31 December 2019. Available data elements included inpatient and outpatient medical claims, pharmacy claims, laboratory claims and sociodemographic information. We also used publicly available data from the ONTT (, the largest randomised, controlled clinical trial of glucocorticoids for ON.8 This study was deemed ‘Not Regulated’ by the University of Michigan Institutional Review Board because it involves only coded private information that cannot be linked to a specific individual by the investigators directly or indirectly through a coding system.

Patient involvement

Patients were not involved in the design, or conduct, or reporting, or dissemination plans of our research.

Study population

Patients were included if they were 18 years or older, had at least one inpatient or outpatient claim for ON as a primary or secondary diagnosis, and were continuously enrolled for 6 months prior to and following the ON diagnosis. ON was identified using the International Classification of Diseases-9 and 10 (ICD-9/10) codes: 337.30, 377.32, 377.39, H46.1x, H46.8X and H46.9x. The first claim date was used as the diagnosis date. Claims within 180 days of the original claim were considered potentially related.9 We excluded patients under 18 years old at diagnosis because treatment and AEs for paediatric ON may be different from adults. Patients with any diagnosis of malignancy, solid organ or bone marrow transplants were excluded because they may be at higher risk for an AE as a result of the underlying disease (online supplemental eTable 1 for all administrative codes). Patients with any glucocorticoid pharmacy or infusion claim in the 6 months prior to their ON diagnosis were excluded because they may be receiving frequent glucocorticoids related to another medical condition. Patients without a pharmacy claim during their entire enrolment period were considered as not having prescription coverage and excluded (figure 1).

Supplemental material

Figure 1
Figure 1

Flow diagram of cohort selection. ON, optic neuritis.


We identified glucocorticoids using the National Drug Code (NDC), generic name or proprietary name of the drug in the pharmacy claims. Additionally, Healthcare Common Procedure Coding System (HCPCS) codes were used to identify glucocorticoid infusions/injections (online supplemental eTable 1). Topical, inhaled, intranasal and intra-articular glucocorticoids were excluded. Fludrocortisone was also excluded given that it is used for chronic conditions (eg, panhypopituitarism). Glucocorticoid infusions/injections administered on the same day as a rituximab, infliximab, ocrelizumab or alemtuzumab infusion were not considered since glucocorticoids are routinely administered as a premedication. Standard oral glucocorticoid doses for each patient were calculated by converting the glucocorticoid formulation into daily prednisone-equivalent doses (online supplemental eTable 2).10 11 For oral glucocorticoids, the duration of glucocorticoid use was calculated using the ‘days’ supply’ variable within the pharmacy claim file. The dose of glucocorticoid infusion/injections cannot be reliably calculated because of the way injections/infusions are recorded within the data set.


Potential AEs were identified through review of the medical literature and analysed overall, by severity and subtype.5–7 AE severity was determined based on potential for hospitalisation or death. AEs classified as mild included headache, mood disorders, sleep disorders, nausea, dizziness, hyperglycaemia and ocular hypertension. AEs classifies as severe included acute pancreatitis, osteonecrosis, heart failure, gastrointestinal perforation or bleeding, fracture, sepsis and venous thromboembolism. The corresponding ICD9/10 codes were ascertained by searching the Healthcare Cost and Utilisation Project Clinical Classification Software categories for relevant codes.12 The Clinical Classification Software collapses the ICD9/10 diagnostic and procedure codes into clinically meaningful categories or specific conditions. Outcomes were assessed within 90 days of the prescription date. For those patients who did not receive glucocorticoids, AEs were ascertained within 90 days of the ON visit date. To address legacy effects within individual patients, any specific AE occurring in the 6 months prior to an ON claim that also occurred in the 90 days following a glucocorticoid prescription was not counted.

Statistical analysis

We used a high-dimensional propensity score approach to address confounding due to sociodemographic factors and comorbidities.13 In contrast to traditional propensity score approaches, we obtained the 100 most common claims in each group: diagnoses, pharmacy claims and procedures observed in our cohort and how frequently they appeared in the records of our patients with ON (online supplemental eTable 3). For each, we computed the quantiles of the number of occurrences per patient. We constructed three indicator variables for each patient: presence of one or more of these records, more than the median number of these records and more than the 75th percentile of these records. We included patient’s age, race, sex and geographic region along with the above indicator variables to obtain propensity scores using multivariable logistic regression. Patients were grouped into quartiles based on their derived propensity scores. Next, we fit a logistic regression model relating a patient’s glucocorticoid use to their probability of experiencing different AEs, controlling for the patient’s propensity score quartile to account for confounding. Each AE outcome was modelled separately. We used generalised estimating equations to account for repeated visits and obtain robust SEs for inference. We obtained population-averaged effects by assuming an autoregressive structure for the working correlation between visits from the same patient.

We performed three additional sensitivity analyses. First, we assessed AEs occurring within 30 days of glucocorticoid administration or ON claim date for those not receiving glucocorticoids. Second, we restricted the cohort to patients with one inpatient or three outpatient ON claims within 60 days of the first diagnosis (PPV=100%) to assess if our results were robust to differences in coding accuracy.14 Third, we assessed AEs among patients receiving high-dose glucocorticoids (≥100 mg prednisone equivalent or injection/infusion (including corticotropin injection) ≥1 day) by adding an interaction effect between glucocorticoid prescription and a dose indicator as oral glucocorticoids are frequently administered at doses that are bioequivalent to injections/infusions. Last, we used a clinician-driven traditional propensity score model, rather than a high dimensional approach, to assess for confounding by indication. Patients’ propensity scores were calculated based on age, sex, race, education level, census division, presence of MS or sarcoidosis at the first ON diagnosis. Neuromyelitis optica, lupus and Sjogren’s disease were not included due to very small, or zero patients seen for these conditions in the 180 days prior to their analysed ON claim.

All statistical analyses were conducted using R V.4.0.3 (released October 2020).


From 2005 to 2019, there were 48 216 medical claims for ON that met the inclusion criteria (figure 1); 19 635 claims were excluded because they occurred within 180 days of another claim for ON; 11 033 claims were removed due to a glucocorticoid pharmacy claim or infusion/injection in the year leading up to the ON claim; 140 claims were removed because the patient was under 18 years of age. The final data set consisted of 17 404 claims from 14 311 unique patients; 434 of these were inpatient ON diagnoses and 16 970 outpatient ON diagnoses.

The median age (IQR) of our cohort at first visit was 48 (37,60) years; 2.9% (n=369) were Asian, 11.0% (n=1399) were Black, 7.9% (n=1007) were Hispanic and 78.2% (n=9940) were White (table 1). Female patients accounted for 66.3% (n=9481). There were 1953 patients (13.6%) who had at least two analysed eligible claims with an ON diagnosis. The maximum number of eligible claims for a single patient was 12.

Table 1

Demographic characteristics of patients with optic neuritis at first diagnosis by glucocorticoid prescription status

Of the 17 404 ON claims analysed, 15.7% (n=2733) had a prescribed glucocorticoid within 30 days of the claim. High-dose glucocorticoids accounted for 40.4% (n=1106) of those prescribed. The median time (IQR) from diagnosis to glucocorticoid prescription or injection/infusion was 31 8 days. The median (IQR) duration of glucocorticoid use was 10 (6,20) days. The median (IQR) daily dose for oral glucocorticoids was 40 (10,60) mg prednisone-equivalents. The dose of infused/injected glucocorticoids could not be determined.

The frequencies of any AE and severe AEs were higher among patients prescribed glucocorticoid versus no glucocorticoids (any AEs: n=437/2733 (16.0%) vs n=1784/14 671 (12.2%]), +3.8% (95% CI: 2.1 to 5.6%)) absolute difference, +29.8% relative difference; severe AEs: n=72/2733 (2.6%) vs n=273/14 671 (1.9%), +0.8% (95% CI: 0.02% to 1.53%) absolute difference, +31.1% relative difference). Among those patients prescribed glucocorticoids, the severity of the AE was mild in 13.3% (n=365) and severe in 2.6% (n=72). The severity of the AEs in those patients not prescribed glucocorticoids was mild in 10.2% (n=1511) and severe in 1.9% (n=273). Frequency of AEs by type and glucocorticoid prescription are detailed in table 2.

Table 2

Frequency of adverse events following an optic neuritis claim by glucocorticoid prescription status

The propensity score adjusted odds of any AE were significantly higher among patients treated with glucocorticoids compared with those patients who were not prescribed a glucocorticoid (OR 1.33 (95% CI: 1.18 to 1.50); p<0.001) (figure 2, table 3). The glucocorticoid group had significantly higher odds of a severe AE within 90 days compared with the non-glucocorticoid group (OR 1.82 (95% CI: 1.37 to 2.41); p<0.001). The odds of mood disorder (OR 1.41 (95% CI: 1.10 to 1.80); p=0.01), hyperglycaemia (OR 4.22 (95% CI: 2.74 to 6.50); p<0.001), sleep disorder (OR 1.45 (95% CI: 1.06 to 1.97); p=0.02), heart failure (OR 3.35 (95% CI: 1.38 to 8.17); p=0.01), sepsis (OR 2.41 (95% CI: 1.22 to 4.74); p=0.01) and venous thromboembolism (OR 2.93 (95% CI: 1.63 to 5.28); p<0.001) were significantly higher among patients prescribed glucocorticoids compared with those patients not prescribed glucocorticoids.

Figure 2
Figure 2

Forest plot of adjusted odds with 95% CIs of adverse events among patients treated with glucocorticoids compared with those who did not receive glucocorticoids. Stricter definition=sensitivity analysis restricting the cohort to patients who had one inpatient claim or three E/M visits with an optic neuritis claim within 60 days of the first diagnosis. Time restricted=sensitivity analysis considering adverse events only occurring within 30 days of glucocorticoid administration or the ON claim date for those not receiving glucocorticoids. ON, optic neuritis.

Table 3

Adjusted odds of adverse events among patients with optic neuritis prescribed glucocorticoids compared with patients not prescribed glucocorticoids

Sensitivity analysis

When the observation period for AE was limited to 30 days or a more specific definition of ON was used, the odds of any AE and severe AE remained higher in the glucocorticoid group than the non-glucocorticoid group (table 3). The odds of any AE were not significantly different between those prescribed high versus low dose glucocorticoids. However, the odds of severe AEs were lower among patients prescribed high-dose compared with low-dose glucocorticoids (OR 0.57 (95% CI: 0.34 to 0.96); p=0.03) (online supplemental eTable 4). The propensity score adjusted odds of any or severe AEs was functionally unchanged when a traditional propensity score approach was used (online supplemental eTable 5).


In this real-world study of more than 17 000 claims from a large private insurer in 2005–2019, we found that ON patients prescribed glucocorticoids had increased odds of AEs captured within a medical encounter within 90 days compared with ON patients who did not receive glucocorticoids. The absolute differences in any (+3.8% (+29.8% relative difference)) and severe (+0.8% (+31.1% relative difference)) AEs were higher among ON patients prescribed glucocorticoids. Importantly, the increase in frequency was for both mild AEs (eg, mood disorders and hyperglycaemia) and severe AEs (eg, venous thromboembolism). Furthermore, the increase in any and severe AEs remained similar in multiple sensitivity analyses that limited the observation period to 30 days from the glucocorticoid prescription, used a more specific and validated definition of ON, and assessed for confounding by indication using a clinician-directed approach. The median duration of prescription for oral glucocorticoids in our study was only 10 days, suggesting that the AE’s captured are unlikely related to chronic glucocorticoid use. Our data can be used by clinicians alongside clinical trial and other prospective data to counsel patients about the potential risks of glucocorticoids.

To the best of our knowledge, this is the first large-scale study of real-world glucocorticoid AEs among patients with ON. Previously, AEs of short-term glucocorticoid use among patients with ON have been captured primarily within clinical trials and not in the real world. Real-world pharmacovigilance data are important for medical decision making as it is well recognised that all potential side effects of medications cannot be fully captured within clinical trials conducted on relatively small and highly selected patient populations treated under tightly controlled conditions. Our study provides this real-world evidence of glucocorticoid-associated AEs occurring among patients with ON, supports observations from the ONTT and adds new information about other potentially serious AEs, not previously captured. In addition to the two serious AEs captured among patients treated with glucocorticoids in the ONTT, approximately half of all participants treated with glucocorticoids experienced sleep disturbance or stomach upset compared with 20% in the placebo group (NNH=3–4), and 34%–36% experienced a mood disorder compared with 16% of patients in the placebo group (NNH=5) (see online supplemental eTable 6).5 As AEs could only be captured in our study within the context of a medical encounter, our estimates for non-life threatening AEs (mild) are understandably lower and likely underestimate the true frequency. However, our data provide a complimentary estimate of real-world AEs (NNH=32) and severe AEs (NNH=143) within 90 days that are severe enough to result in an inpatient or outpatient medical visit or reported within the context of a medical visit.

It is important to consider that for patients with the most common ON subtypes (idiopathic ON, MS-associated ON), there are no long-term benefits to glucocorticoid treatment. In comparison to the risks of glucocorticoids, the average absolute short-term improvement in visual function among patients with the most common types of ON treated with glucocorticoids is lower than the absolute risk (+2.9% (95% CI: −3.95 to 9.88)).4 To place into context the short-lived, average improvement in high-contrast visual acuity, about 34 patients with ON need to be treated for a single person to experience the average benefit of a 10-letter improvement in high-contrast visual acuity at 15 days or return to Snellen visual acuity of 20/20 or better in patients with less severe vision loss. Importantly, there is no benefit at 30 days, 6 months or 1 year (online supplemental eTable 6).2 4 For some select patients (eg, commercial drivers), these potential marginal improvements may outweigh the potential harms; however, it is unclear if this improvement translates into improved functioning for the average person. It is important to note that approximately 10% of patients with ON have a glucocorticoid-responsive subtype (eg, neuromyelitis optica spectrum disorder, myelin oligodendrocyte glycoprotein associated disease)3 and estimated benefit data from the ONTT should not be applied as few patients with glucocorticoid-responsive subtypes were enrolled.15

Our findings also expand on two prior population-based studies in adults evaluating the risk of short-term glucocorticoids. Waljee et al6 found that among a population of 1.5 million US adults aged 18–64 years, the risk of sepsis, venous thromboembolism and fracture were 2-to-5 fold higher than background rates within 30 days of oral glucocorticoid administration. Similar to our study, the median prednisone-equivalent dose was 20 mg daily with 23% receiving 40 mg or more per day. The median treatment duration was 6 days. Similar findings were observed among a cohort of 2.6 million adults in Taiwan who received a single burst of glucocorticoids.7 While many of the AEs we observed among ON patients were similar, such as sepsis, we did not observe an increase in some AEs previously observed in prior studies. For some AEs, few cases were captured limiting our study power to detect a difference between groups. For others, such as venous thromboembolism, where case capture was adequate, differences in our population and underlying disease may result in distinctive risks.

Our study has important limitations. Because we used private insurance claims, our findings may not be generalisable to those who are uninsured. Likewise, the majority of patients in our study were White, had posthigh school education, and had a high household income which is not representative of the US population and could have biased our results. Although we did not observe differences by glucocorticoid dose for any AE, the odds of a severe AE was lower among patients receiving high-dose versus low-dose glucocorticoids and may suggest a relationship between severity of AEs, glucocorticoid dose and duration of use. However, these results should be interpreted cautiously given the constraints of the data set, in which determining dosage or changes in dosing over time is limited. Furthermore, the diagnosis of ON may be miscoded in the data set. However, our findings were unchanged in a sensitivity analysis using an ON definition with PPV=100% and our aim was to assess AEs of glucocorticoids in a manner that reflects real-world practice, where diagnostic certainty may be limited in the acute phase. Therefore, we purposely used a less specific definition of ON, which led to our overall lower proportion of glucocorticoid prescriptions compared with the high utilisation reported by clinicians.16 Likewise, because this is a real-world study that included all types of ON, this likely explains why our population is slightly older and had a lower proportion of women overall in comparison to the ONTT, which had strict age criteria. Notably, our demographic data are not dissimilar to other recent epidemiologic studies.3 Our study design assumes that the glucocorticoid was prescribed in response to the ON diagnosis given the proximity of the prescription and ON claim. Although the approach we took is similar to prior studies, it is possible that the glucocorticoid was prescribed for another reason that was not fully documented in the data set. Last, our study accounts for a large number of factors and likely interactions between those factors, but it does not provide direct evidence for how specific factors interact to cause AEs, cannot address residual confounding, nor establish causality between glucocorticoids and AEs. Future work to understand this problem may inform clinicians about which patients should avoid glucocorticoids.

In conclusion, glucocorticoids prescribed in the real world for patients with ON are associated with 30% relative increase in AEs within 90 days, some severe such as VTE. This information should be communicated alongside clinical trial AE data to communicate risks of glucocorticoids. For patients with the most common ON subtypes, in which the improvement in visual functioning with glucocorticoids is marginal and short lived, these data can help support shared decision-making conversations.

Data availability statement

Data may be obtained from a third party and are not publicly available.

Ethics statements

Patient consent for publication

Ethics approval

This study was exempt from review by the University of Michigan Institutional Review Board and informed consent was not required.

This post was originally published on