Identification of Ocular and Systemic Predictors and Development of a Prediction Model in Ocular Myasthenia Gravis

Abstract

Purpose

Predicting patients with ocular myasthenia gravis (OMG) achieving minimal manifestation status (MMS) or better is crucial for guiding treatment decisions. Prognostic models are becoming increasingly popular in the medical field due to their high accuracy. This study aimed to develop a clinical tool based on both ocular and systemic factors.

Methods

We reviewed patients from our hospital from 2020 to 2023. Logistic and Least Absolute Shrinkage and Selection Operator (LASSO) regressions were applied to determine key predictors. A logistic nomogram model was constructed based on the selected factors. The model’s performance was rigorously evaluated through internal validation for discrimination, calibration, and clinical utility.

Results

Among 216 patients, 126 individuals were included in the study. A total of 61 patients (48.4%) achieved MMS or better status in the 1-year outcome. Five key variables were selected to conduct the prognostic model, including ocular deviation angle, duration, antibody, corticosteroid therapy, and thyroid ultrasound. The area under the curve (AUC) was 0.889 (95% confidence interval [CI] = 0.832–0.947), demonstrating a good performance. The calibration curves also showed excellent agreement between predicted and observed probabilities in both training and validation cohorts.

Conclusions

This study provides a clinically applicable predictive model, which accurately identifies patients with OMG likely to achieve MMS or better status within 1 year. The model is predictive of individual patients with great performance, thereby increasing clinical confidence and utility.

Translational Relevance

This study translates specific ocular examination findings into a practical prognostic tool. The developed model directly assists clinicians in identifying patients with OMG with a high probability of achieving remission within 1 year, thereby informing personalized management strategies.

Introduction

Myasthenia gravis (MG) is an autoimmune disease characterized by pathological changes at the neuromuscular junction, resulting in progressive weakness in ocular, facial, and limb muscles that worsens with fatigue.^1,2 Ocular myasthenia gravis (OMG), a subtype of MG, is mainly characterized by ptosis and diplopia. Approximately 50% of patients with OMG manifest antibodies targeting neuromuscular receptors, with a dominant acetylcholine receptor (AChR) antibody phenotype and a relatively low prevalence of antibodies against muscle-specific kinase (MuSK) or low-density lipoprotein receptor-associated proteins.³

Appropriate therapeutic intervention and clinical management of OMG are crucial to alleviating visual impairment and improving patients’ quality of life. As outlined in the Clinical Research Guidelines of the Myasthenia Gravis Foundation of America (MGFA), minimal manifestation status (MMS) serves as a standardized index for evaluating clinical status and indicates a principal outcome measure in patients with MG.^4,5 Most studies on the prognosis of patients with OMG have focused on generalizing to generalized myasthenia gravis (GMG).^6,7 Limited evidence has been conducted on treatment-induced remission status in OMG, including complete stable remission (CSR) or MMS.^8,9 The appraisal of MMS as a therapeutic endpoint is critical for optimizing treatment strategies. Furthermore, prior prognostic research has largely relied on broad clinical descriptors, whereas quantitative, standardized ophthalmic examination-based parameters are still insufficiently characterized, such as graded ptosis severity, comprehensive extraocular motility limitation, and the prism-measured ocular deviation angle. Given the clinical manifestation of OMG into consideration, comprehensive data on ophthalmic parameters are essential for accurate prognostic analysis.

Nomogram is a graphical tool for synthesizing multivariable data into individualized risk probabilities, and has been widely used in the clinical area, including ophthalmology and neurology.^10,11 Accurate individual recurrence prognosis is crucial for tailoring treatment strategies in OMG. Although several factors have been associated with OMG outcomes, there is a lack of integrated tools that quantify the combined effect of these factors to provide an individualized risk estimate. Prediction models could combine multiple predictors to estimate the probability of MMS for patients with OMG.

Therefore, this study aims to develop and internally validate a multivariable prediction model for achieving MMS in patients with OMG, and to present it as an easy-to-use nomogram for clinical application. The proposed model will integrate the synthesis of ocular and systemic parameters to establish a robust prognostic tool for predicting OMG outcomes at onset.

Methods

Study Subjects

Retrospective data were collected from 216 patients with OMG who were diagnosed first in the ophthalmology clinic of the First Affiliated Hospital of Jinan University from January 2020 to December 2023. We diagnosed OMG based on clinical signs of fluctuating ophthalmic muscle weakness, supported by positive findings in at least one of three diagnostic evaluations: serum antibody test, electromyography (EMG), or neostigmine test.¹² All the patients were included and followed up for at least 12 months. Our study protocol was approved by the First Affiliated Hospital of Jinan University Ethics Committee (Approval No.: KY-2021-131). Because this study analyzed de-identified information retrieved from the medical record system, the written informed consent was waived for the participants. The study design and workflow are illustrated in Figure 1.

Figure 1.

Workflow of the study.

Inclusion Criteria and Exclusion Criteria

Inclusion criteria for our study: (1) patients diagnosed with OMG for the first time in our ophthalmological department; and (2) a minimum follow-up period of 12 months after diagnosis. The exclusion criteria were the following: (1) progression to GMG at the beginning or within the first 3 months of the diagnosis of OMG; (2) coexisting pathologies underlying mingling ocular manifestations, such as thyroid eye disease (TED), congenital MG syndrome, congenital ptosis, etc.; (3) history of strabismus surgical intervention; (4) current use of medications known to exacerbate MG or interact adversely with corticosteroids (e.g., specific antiviral agents, aminoglycosides, or systemic immunosuppressants) prior to the initial consultation; and (5) previous surgical therapy of thymectomy or thymus radiation therapy.

Clinical outcomes were evaluated according to the MGFA. The MMS was determined through an overall clinical examination. Patients who have not received any treatment within the past year were considered MMS-0. Those receiving immunosuppressive therapy without additional symptomatic treatment were classified as MMS-1. The MMS-2 classification is designated for patients sustained on low-dose pyridostigmine monotherapy (<120 mg/day) for ≥12 months, whereas MMS-3 is assigned to those patients requiring both symptomatic treatment and immunosuppressive therapy during the observation stage. Exclusion of GMG within 1 month of diagnosing OMG strictly adheres to the MGFA clinical classification. Through comprehensive clinical history collection and targeted neurological examinations, systemic involvement is rigorously excluded to confirm the complete absence of facial, limb, or respiratory muscle weakness. For patients whose clinical symptoms are limited to ocular muscles, chest computed tomography (CT) is used to assess thymic abnormalities. Those with thymic abnormalities undergo respiratory neurological examinations and non-ocular electromyography to further rule out GMG. Oral corticosteroids are initiated at a low dose of 5 mg daily, with monthly assessments of symptom improvement. If clinical improvement is insufficient, the dose is gradually increased up to 40 mg daily. All patients received pyridostigmine therapy according to international treatment guidelines.¹³

Data Collection

Comprehensive clinical data were collected from all the patients, including demographic characteristics (age, sex, and duration), systemic examinations (antibody, repetitive nerve stimulation [RNS], ice test, neostigmine trial, lymphocyte, thymus CT, thyroid function, and thyroid ultrasound), ophthalmological characteristics (OMG type, type of strabismus, ocular deviation angle, eye movement, extraocular muscle paralysis, type of ptosis, ptosis symmetry, and eyelid height) and treatments (corticosteroid therapy, traditional Chinese medicine, and immunosuppressive therapy). Duration denoted the time from the onset of disease symptoms to the diagnosis of the patient. Thyroid ultrasound abnormalities were defined as nodules, cysts, calcifications, or diffuse changes with a Chinese thyroid imaging reports and data systems (C-TIRADS) score ≥3. These included features such as enlargement, atrophy, hypoechoic patterns, and heterogeneous echogenicity.¹² OMG subtypes were classified according to age of occurrence: childhood-onset OMG (<18 years; C-OMG), early-onset OMG ≥18 years and ≤50 years (E-OMG), and late-onset OMG >50 years (L-OMG). The angle measurement refers to the angle of ocular misalignment or angle of strabismus when the patient’s eyes are in the primary position. We used a red Maddox rod for qualitative assessment of the direction of ocular diplopia. Then, we measured the angle of ocular deviation quantitatively through the red-glass test combined with prism neutralization. Prisms were incrementally added until the red and white images observed by the patient completely overlapped, with the final measurement recorded in prism diopter (PD). The ocular deviation angle was originally measured in PD, and for statistical analysis was expressed in 10-PD units.

Our analytical framework included 22 candidate characteristics, which underwent rigorous selection through the Least Absolute Shrinkage and Selection Operator (LASSO) regression and multivariate logistic regression. A 10-fold cross-validated LASSO was applied to select the optimal regularization parameter (λ). The final value of λ is established using the 1 standard error (1-SE) rule (lambda.1se), and the variables that correspond to nonzero coefficients are selected as candidate predictors.

Model Development

We derived the model based on the significant variables from the univariable logistic regression and LASSO regression. All the selected factors were assessed for multicollinearity using the variance inflation factor (VIF) analysis, utilizing a conservative threshold of VIF ≤5 to exclude variables illustrating excessive collinearity.

Model Verification

We validated the nomogram through a bootstrap 1000-replicated sampling internal validation. We adopted multiple analysis metrics to assess the performance of the model on both original and internal cohorts: receiver operating curves (ROCs), calibration curves, and decision curve analysis (DCA), respectively. Furthermore, the predictive capacity of the model was deemed superior when the bias-corrected curve was in closer alignment with the ideal curve. To gain further insight into the sensitivity and specificity of the model, the decision curve and impact curve were plotted.

Statistical Analysis

All statistical analyses and data visualizations were executed using R software version 4.4.2. Continuous data were usually reported as mean ± standard deviation or median (interquartile spacing),¹² whereas categorical data were presented as frequencies (percentages). Differences between groups were compared using chi-square tests for categorical variables and independent samples t-tests or nonparametric tests for continuous variables. The “pROC” package was used for ROC analysis, and the “rmda” package for decision curve analysis. A 2-sided P value < 0.05 was considered statistically significant.

Results

Baseline Characteristics

Among the 216 patients, 4 had associated surgery histories, and 6 individuals generalized within 1 month after diagnosed. After screening based on the criteria, a total of 126 patients with OMG were included in the study. A total of 61 patients reached MMS or better status (see Fig. 1). After excluding patients who generalized to GMG within 3 months of diagnosis, we observed that 2 patients later progressed to GMG during follow-up. Demographic and clinical characteristics of patients achieving, compared to not achieving MMS or better, are shown in Table 1. Duration, immunosuppressive therapy, antibody, corticosteroid therapy, ocular deviation angle, and thyroid ultrasound revealed significant differences between the two groups (P < 0.05). The ocular deviation angle is presented in 10-PD units, and a value of 1.0 corresponds to 10 PD in the original clinical measurement.

Table 1.

Basic Characteristics of the Patients in the Cohorts

Characteristics	Total (n = 126)	No MMS (n = 65)	MMS (n = 61)	P Value
Sex, n (%)				0.208
Male	61 (48.4)	26 (40.0)	35 (57.4)
Female	65 (51.6)	39 (60.0)	26 (42.6)
Age, y	31.5 (20.0–53.0)	34.0 (20.0–55.0)	28.0 (19.0–46.0)	0.164
OMG type, n (%)				0.212
C-OMG	29 (23.0)	14 (21.5)	15 (24.6)
E-OMG	61 (48.4)	28 (43.1)	33 (54.1)
L-OMG	36 (28.6)	23 (35.4)	13 (21.3)
Duration, y	0.5 (0.2–2.0)	1.0 (0.2–2.0)	0.2 (0.1–1.0)	0.001
Corticosteroid therapy, n (%)				0.006
No	84 (66.7)	36 (55.4)	48 (78.7)
Yes	42 (33.3)	29 (44.6)	13 (21.3)
Chinese medicine, n (%)				0.133
No	100 (79.4)	55 (84.6)	45 (73.8)
Yes	26 (20.6)	10 (15.4)	16 (26.2)
Immunosuppressive therapy, n (%)				0.017
No	98 (77.8)	45 (69.2)	53 (86.9)
Yes	28 (22.2)	20 (30.8)	8 (13.1)
Antibody, n (%)				<0.001
Negative	92 (73.0)	38 (58.5)	54 (88.5)
Positive	34 (27.0)	27 (41.5)	7 (11.5)
RNS, n (%)				0.087
Abnormal	77 (61.1)	37 (56.9)	40 (65.6)
Normal	17 (13.5)	13 (20.0)	4 (6.56)
Unknown	32 (25.4)	15 (23.1)	17 (27.9)
Ice test, n (%)				0.745
Positive	60 (47.6)	29 (44.6)	31 (50.8)
Negative	25 (19.8)	13 (20.0)	12 (19.7)
Unknown	41 (32.5)	23 (35.4)	18 (29.5)
Neostigmine trial, n (%)				0.313
Negative	28 (22.2)	18 (27.7)	10 (16.4)
Positive	73 (57.9)	35 (53.8)	38 (62.3)
Unknown	25 (19.8)	12 (18.5)	13 (21.3)
Lymphocyte, n (%)				0.811
Normal	35 (27.8)	17 (26.2)	18 (29.5)
Abnormal	59 (46.8)	30 (46.2)	29 (47.5)
Unknown	32 (25.4)	18 (27.7)	14 (23.0)
Thymus CT, n (%)				0.056
Normal	71 (56.3)	31 (47.7)	40 (65.6)
Abnormal	52 (41.3)	33 (50.8)	19 (31.1)
Unknown	3 (2.38)	1 (1.54)	2 (3.28)
Thyroid function, n (%)				0.910
Normal	75 (59.5)	39 (60.0)	36 (59.0)
Abnormal	51 (40.5)	26 (40.0)	25 (41.0)
Thyroid ultrasound, n (%)				0.021
Normal	47 (37.3)	18 (27.7)	29 (47.5)
Abnormal	79 (62.7)	47 (72.3)	32 (52.5)
Type of strabismus, n (%)				0.313
None	34 (27.0)	13 (20.0)	21 (34.4)
Horizontal	45 (35.7)	26 (40.0)	19 (31.1)
Vertical	24 (19.0)	14 (21.5)	10 (16.4)
Mixed	23 (18.3)	12 (18.5)	11 (18.0)
Ocular deviation angle (10 PD)	0.5 (0–2.0)	1.0 (0–2.5)	0.2 (0–0.8)	0.002
Eye movement, n (%)				0.394
Normal	55 (43.7)	26 (40.0)	29 (47.5)
Abnormal	71 (56.3)	39 (60.0)	32 (52.5)
Extraocular muscle paralysis, n (%)				0.169
Single	46 (36.5)	25 (38.5)	21 (34.4)
Numerous	48 (38.1)	28 (43.1)	20 (32.8)
None	32 (25.4)	12 (18.5)	20 (32.8)
Type of ptosis, n (%)				0.864
Left	49 (38.9)	24 (36.9)	25 (41.0)
Right	16 (12.7)	9 (13.8)	7 (11.5)
Both	61 (48.4)	32 (49.2)	29 (47.5)
Ptosis symmetry, n (%)				0.999
Yes	10 (7.94)	5 (7.69)	5 (8.20)
No	55 (43.7)	28 (43.1)	27 (44.3)
None	61 (48.4)	32 (49.2)	29 (47.5)
Lid height, mm	1.5 (0–6.0)	5.0 (0–9.0)	6.1 (0–9.0)	0.445
Generalized, n (%)				0.504
Yes	2 (1.6)	2 (3.1)	0 (0)
No	124 (98.4)	63 (96.9)	61 (100.0)

MMS, minimal manifestation status; OMG, ocular myasthenia gravis; RNS, repetitive nerve stimulation.

Risk Factors

Among all the included features, LASSO regression in the training cohort was shown in Figure 2. This method chose the cleanest model with five variables as follows: duration, ocular deviation angle, antibody, thyroid ultrasound, and corticosteroid therapy. In addition, we utilized the multivariate logistic regression and showed that antibody, ocular deviation angle, duration, thyroid ultrasound, and corticosteroid therapy were also significant variables (Table 2). All the factors were risk factors except corticosteroid therapy. All the VIF values of the selected 5 factors were under 5.

Figure 2.

(A) LASSO Coefficient distribution map-LASSO coefficient distribution of all variables. (B) Variables determined by LASSO analysis (n = 5). LASSO, Least Absolute Shrinkage and Selection Operator.

Table 2.

Multivariable Logistic Regression Analyses of Factors Associated With Achieving MMS

Variables	OR (95% CI)	P Value
Duration, per y	0.80 (0.73–0.99)	0.009
Corticosteroid therapy	3.03 (1.04–10.09)	0.045
Antibody	0.15 (0.04–0.41)	<0.001
Thyroid ultrasound	0.37 (0.13–0.88)	0.027
Ocular deviation angle, per 10 PD	0.71 (0.55–0.90)	0.006

OR, odds ratio; CI, confidence interval.

Model Performance

We performed a logistic regression model to predict the outcome, which patients with OMG achieved MMS or better status after 1 year of follow-up. The model was integrated into the nomogram (Fig. 3). We also constructed a convenient online dynamic nomogram website, available at https://ocularyrc.shinyapps.io/OMG-nomogram/. the prediction value could be calculated immediately by inputting the numerical value of each variable.

Figure 3.

Nomogram for the prediction of patients with OMG reaching MMS or better after 1 year.

The ROC curves of the models in the training and validation sets were displayed in Figure 4. The training cohort predicts MMS with an area under the curve (AUC) 0.889 (95% confidence interval [CI] = 0.832–0.947), and the validation cohort showed 0.813 (95% CI = 0.739–0.887). The calibration curves also appeared to be almost diagonal, and the risk score provides a greater benefit compared to the extreme curve according to the DCA curves in the two sets (see Figs. 4E, 4F). In addition, Table 3 presents several evaluation indicators, including sensitivity, specificity, and the Youden index. The sensitivity was 82.3% and the specificity was 91.4%. All the results indicated that the predictive model exhibited fair performance.

Figure 4.

(A) ROCs of the training cohort and (B) validation cohort. The calibration curve of (C) the training cohort and (D) the validation cohort. The DCA curve of (E) the training cohort and (F) the validation cohort.

Table 3.

The Ability of the Training and Validation Cohorts

Cohort	Specificity	Sensitivity	Youden Index
Training	0.82	0.91	0.74
Validation	0.84	0.68	0.51

Discussion

This study developed an OMG outcome assessment method using a predictive nomogram to predict the MMS or better status after 1 year of follow-up. The performance of the model is assessed through the training and validation cohorts. The AUC, specificity, sensitivity, and other indicators were used to evaluate the performance of the nomogram, and showed that the model has a high prediction ability.

MMS is a clinical status description of the post-intervention state of patients with MG, which was first proposed by the MGFA in 2000.⁸ Recently, the status has been considered as one of the main outcomes in patients with MG.^2,13 However, most studies performed to predict the progress of OMG were focused on generalizing to GMG. As MMS is an important landmark in the treatment of MG, we focused on the MMS or better status of patients with OMG. Li et al. performed a study about factors affecting MMS among MG and demonstrated that isolated ocular involvement is beneficial for MMS induction.¹⁴ In Ariatti's research, 39 of 45 patients (excluded generalizing to GMG) received MMS after a 3-year follow-up.³ In addition, another Thailand study showed 66.7% (58 of 81) of patients followed up for 1 year or more were able to achieve MMS.¹⁵ Our MMS or better status rate of 48.4% was comparable to the findings of these studies, as we performed follow-up in 1 year and had a larger population.

The major strength of our study was the included detailed ocular examination results for the first time and have found the ocular deviation angle to be a predictor of outcome in patients with OMG. The ocular manifestations of patients with OMG have been noticed in prior studies, especially the severity of strabismus or diplopia, and ptosis. Strabismus surgery has been used to treat patients with OMG who have a larger angle of deviation at presentation, and improved the prognosis of refractory patients.^6,16 Whereas our study found that the ocular deviation angle was equally related to the prognosis of nonsurgical treated patients. This finding added to the correlation between the ocular deviation angle and outcome in patients with OMG. In addition, Indian experts conducted research on OMG treatment outcomes and included different eye manifestations. Patients presented with ptosis or diplopia, or both, showed no significant difference.¹⁷ By contrast, consistent with prior findings, the presence of ptosis versus diplopia did not differentiate outcomes in our cohort, suggesting that quantitative assessment of ocular deviation may be more informative than symptom category alone.

Consistent with previous studies, anti-AChR Ab positivity was a significant factor in the prognosis of patients with OMG with adverse outcomes.^4,15,18 In addition, systemic corticosteroid therapy was also a significant treatment for patients with OMG, and was considered as one indicator of the outcome of patients with OMG. Kessi et al. found that children with OMG who received glucocorticoids within 6 months of diagnosis had a greater proportion of optimal outcomes.¹⁹ For suppressing progression to GMG in patients with OMG, long-term low-dose prednisolone is effective.²⁰ Another research that focused on the MMS status in Japan also highlighted that patients with OMG early received corticosteroids within 255 days from onset were more likely to achieve this status.²¹ In our study, having corticosteroid therapy throughout the 1-year course of treatment increases the likelihood of obtaining MMS or better status. This finding supports the clinical value of corticosteroid therapy in appropriate patients with OMG.

As an immune-related disorder, thyroid disease is associated with MG. Autoimmune thyroid disorders (AITD), including Graves’ disease and Hashimoto's thyroiditis, was reported to be more likely suffered from patients with MG.²² A meta-analysis also indicated that patients with MG exhibit an elevated risk of thyroid dysfunction and thyroid autoimmune diseases.²³ As ultrasound is the most common and useful method for detecting and diagnosing thyroid diseases, it was also one important evaluation of patients with MG. In Japan, a study showed that 77.2% among patients with MG were found to have an abnormal thyroid ultrasound.²⁴ Previous research only focused on the association between thyroid disease and MG, with limited exploration of thyroid conditions specifically in patients with OMG. Our findings indicate that thyroid status is a significant determinant of outcomes in patients with OMG, distinct from TED. We found that whereas traditional thyroid serological markers showed no significant predictive value, structural abnormalities detected by thyroid ultrasound emerged as an independent risk factor. This discrepancy may relate to ultrasound's sensitivity in detecting early or localized thyroid autoimmunity. Parenchymal changes may precede systemic serological manifestations, making ultrasound findings a potential indicator of underlying immune dysfunction. In the future, we should prioritize comprehensive thyroid examinations for patients with OMG, rather than focusing solely on thyroid function.

Although our study has innovative aspects, several limitations warrant consideration. First, as OMG is a rare disease, the study was conducted at a single center, making it difficult to collect large multi-center datasets. Consequently, the model was trained and validated solely on data from this single population, lacking external validation. Second, we only assessed anti-AChR antibody status and did not include other antibodies (e.g., anti-MuSK), given their low positivity rate in our cohort. This approach may overlook the potential prognostic significance of other antibody profiles. Third, the study focused on short-term treatment outcomes (1-year MMS), lacking longer-term outcome data. Given that many patients with OMG require extended management, future research should incorporate longer follow-up periods. Finally, although the sample size is reasonable for a rare disease, it remains relatively small for developing robust predictive models. Future validation in larger OMG cohorts is essential to enhance the model’s robustness and clinical applicability.

Conclusions

In summary, we developed a prediction model to identify the MMS or better status of patients with OMG after 1 year of treatment. We incorporated exhaustive ocular examination results with prognostic correlations and found the ocular deviation angle to be a significant variable, as well as duration, antibody levels, thyroid ultrasound, and corticosteroid therapy. Using this model to predict patients’ with OMG progress is convenient and can be widely extended to neurological, as well as ophthalmic, healthcare settings. Furthermore, based on prognostic predictions, early efforts can be made to develop therapeutic interventions that are more specific to individual patient disease progression and clinical needs.