Ocular Myasthenia Gravis: Toward a Risk of Generalization Score and Sample Size Calculation for a Randomized Controlled Trial of Disease Modification

The vulnerable brain areas in hypoxic-ischemic encephalopathy (HIE) following systemic hypotension are typically the neocortex, deep cerebral gray nuclei, hippocampus, cerebellum, and the parieto-occipital arterial border zone region. The visual cortex is not commonly recognized as a target in this setting.; ; Single-institution review from 2007 to 2015 of patients who suffered cortical visual loss as an isolated clinical manifestation following systemic hypotension and whose brain imaging showed abnormalities limited to the occipital lobe.; ; Nine patients met inclusion criteria. Visual loss at outset ranged from hand movements to 20/20, but all patients had homonymous field loss at best. In 1 patient, imaging was initially normal but 4 months later showed encephalomalacia. In 2 patients, imaging was initially subtle enough to be recognized as abnormal only when radiologists were advised that cortical visual loss was present.; ; The occipital lobe may be an isolated target in HIE with cortical visual loss as the only clinical manifestation. Imaging performed in the acute period may appear normal or disclose abnormalities subtle enough to be overlooked. Radiologists informed of the clinical manifestations may be more attune to these abnormalities, which will become more apparent months later when occipital volume loss develops.

Original Contribution Ocular Myasthenia Gravis: Toward a Risk of Generalization Score and Sample Size Calculation for a Randomized Controlled Trial of Disease Modification Sui H. Wong, MD, MRCP, Aviva Petrie, CStat, Gordon T. Plant, FRCP, FRCOphth Background: There is currently no prognostic test to determine the risk of developing generalized myasthenia gravis (GMG) risk in patients who first present with ocular disease. Most studies that report risk factors are flawed by the inclusion of patients on immunosuppression, which is likely to reduce the risk. Objective: To create a prognostic score to predict the risk of GMG. Methods: Multicenter retrospective cohort of patients with ocular myasthenia gravis for minimum 3 months, untreated with immunosuppression for minimum 2 years or until GMG onset. Results: One hundred one (57 female) patients were included, with median follow-up of 8.4 years (2-42) from disease onset. Thirty-one developed GMG at median of 1.31 years (3.5 months-20.2 years); 19 occurred within 2 years. Univariable logistic regression analysis produced 3 significant predictors (P , 0.10), adjusted odds ratios in a multivariable logistic model (x2 P = 0.01) with multiple imputations for missing data: seropositivity, 5.64 (95% CI, 1.45-21.97); presence of 1 or more comorbidities including autoimmune disorders, 6.49 (95% CI, 0.78-53.90); thymic hyperplasia, 5.41 (95% CI, 0.39-75.43). Prognostic score was derived from the coefficients of the logistic model: sum of the points (1 point for the presence of each of the above predictive factors), classified "low risk" if #1 and "high risk" if $2. Predicted probabilities were 0.07 (SD, 0.03) for low risk and 0.39 (SD, 0.09) for high risk. Negative predictive value was 91% (95% CI, 79-98), positive predictive Department of Neuro-ophthalmology (SHW, GTP), Moorfields Eye Hospital, London, United Kingdom; Medical Eye Unit, St Thomas' Hospital (SHW, GTP), London, United Kingdom; Department of Neuro-ophthalmology (SHW, GTP), National Hospital for Neurology and Neurosurgery, London, United Kingdom; and Biostatistics Unit (AP), University College London Eastman Dental Institute, London, United Kingdom. The authors report no conflicts of interest. Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the full text and PDF versions of this article on the journal's Web site (www. jneuro-ophthalmology.com). Address correspondence to Sui H. Wong, MRCP, Department of Neuroophthalmology, Moorfields Eye Hospital, 162 City Road, London EC1V 2PD, United Kingdom; E-mail: [e-mail] 252 value was 38% (95% CI, 23-54), sensitivity was 79% (95% CI, 54-94), specificity was 63% (95% CI, 50-74), area under receiver operating characteristic curve was 0.74 (95% CI, 0.64-0.85). Conclusions: In this preliminary study, we have shown by proof of principle that it is possible to stratify risk of GMG: an approach that may allow us to better counsel patients at diagnosis, complement decision-making, and move us toward addressing the question of modifying GMG risk in high-risk patients. Furthermore, the effect of comorbidities is novel and demands further elucidation. Journal of Neuro-Ophthalmology 2016;36:252-258 doi: 10.1097/WNO.0000000000000350 © 2016 by North American Neuro-Ophthalmology Society A large proportion (50%-85%) of patients with autoimmune myasthenia gravis present with only ocular symptoms of diplopia, ptosis, or both (1-3). This is designated as ocular myasthenia gravis (OMG). Within 2 years of onset, 30%-80% develop weakness in the limbs, bulbar, or respiratory muscles, that is, generalized myasthenia gravis (GMG) (1,3-5). A current controversy is whether the risk of GMG can be modified by early immunosuppression; this issue is as yet unresolved because most of the evidence comes from retrospective case series (6). Due to the associated risks of immunosuppression, current clinical guidelines in Europe (7) advise reserving this as second-line treatment, after symptomatic treatment first (e.g., with anticholinesterase inhibitors or prisms). This question of early immunosuppression can be definitively addressed with a prospective randomized controlled trial (RCT). However, we put forward the case that before such a trial can be ethically performed, we need a reliable method of identifying suitable patients, that is, patients who are likely to develop GMG. Currently, there is no test to predict risk of conversion to GMG. Reported risk factors include older age of onset (1,8), seropositivity (9), abnormal repetitive accessory nerve Wong et al: J Neuro-Ophthalmol 2016; 36: 252-258 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution stimulation (5), and severity of symptoms (5). However, most studies included immunosuppressed patients who are likely to influence the outcomes with regard to generalization (6). For the purposes of this study, we have proposed, from clinical experience and theoretical considerations, further potential patient risk profiles for developing GMG. This preliminary study aims to show by proof of principle, using a retrospective cohort, that it is possible to create a prognostic test score to predict the risk of developing GMG. METHODS Study Sample and Method for Data Collection A multicenter retrospective cohort study by review of patient records between May 2013 and July 2014 of patients who first presented to our neuro-ophthalmology service between January 1993 and March 2013; only patients with a minimum of 2-year follow-up (from the time of ocular symptom onset) were included in the study. Patients were seen in our neuro-ophthalmology clinics at Moorfields Eye Hospital, the National Hospital for Neurology and Neurosurgery, and St Thomas' Hospital, London. Patients were identified from clinic lists and previous audits or service reviews, and screened for inclusion and exclusion criteria (See Supplemental Digital Content 1, Table E1, http://links.lww.com/WNO/A188). Variable Definitions Data collection included the following, based on an a priori hypothesis of potential predictors from a literature review and clinical experience. From Literature Review 1. 2. 3. 4. 5. Age onset (1,8). Ethnicity (10). Seropositivity to anti-AChR (5). Seropositivity to anti-MuSK antibodies (11). All anti-AChR tests were performed by the standard radioimmunoprecipitation assay (12) in laboratories in the United Kingdom and historically most sent to the neuroimmunology laboratory in Oxford. Serological tests available throughout the study may have varied; for example, anti-MuSK antibody tests were only available after 2001 and serological tests were not repeated as patients may have subsequently been immunosuppressed or may no longer be under our care. For seropositivity, the analysis for anti-AChR and anti-MuSK was combined (patients had either seropositivity to anti-AChR or anti-MuSK, but not both) because of small number of patients with anti-MuSK. Wong et al: J Neuro-Ophthalmol 2016; 36: 252-258 6. Positive electromyography (EMG), particularly if limb/ trapezius positive (5). From Clinical Experience or Theoretical Considerations 1. Gender: Epidemiological studies on the prevalence of MG show that there is a higher percentage of females with GMG than with OMG (3), suggesting that there may be a difference in the risk of conversion to GMG in females. 2. Presence of comorbidities: That is, other previous illnesses in the medical history as recorded in the notes. We considered this a potential surrogate marker of "resilience," which may influence disease response. 3. Presence of other autoimmune diseases: This is a subgroup of the previous factor ("presence of comorbidities")-we decided to analyze separately comorbid illness due to autoimmunity, which might influence outcome differently from other nonspecific comorbidities (13). 4. Thymic hyperplasia: Detected by radiological observation of thymic hyperplasia. We considered the possibility that thymic hyperplasia may indicate a more severe ongoing autoimmune reaction, which may be a predictor for GMG. 5. Elevated titers of antithyroid antibodies: We and others have observed the frequent elevated antithyroid antibody levels and considered this as a potential predictor for GMG (14). This is not necessarily a subgroup of "presence of autoimmune diseases" other than where such patients have an identified dysthyroid disorder. 6. Ophthalmoparesis: We considered this as a surrogate marker of more severe disease at onset and a possible predictor for GMG. Statistical Analysis The primary outcome measure was GMG (by patient symptom and confirmed by neurologic examination) within 2 years from symptom onset of ocular disease. Statistical analyses were performed with Stata version 13.1 (StataCorp, College Station, TX) as follows. 1. Univariable analyses of possible risk factors as identified in the a priori hypothesis. Tests of categorical variables were by Fisher exact tests and continuous variables by 2sample t-tests or Mann-Whitney tests, as appropriate, depending on whether the data were normally distributed. 2. Multivariable analyses: Potential risk factors from the univariable analyses showing statistical significance at the 10% level were selected for multivariable logistic regression analysis, avoiding combinations of variables that would lead to collinearity. Multiple imputations 253 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution (MIs) were performed for missing data. Tests were performed to investigate potential interactions and confounders, and the effects of adjustment for age, sex, and ethnicity. Likelihood ratio tests were performed to ensure that the final model was as simplified as possible. 3. Goodness of fit was checked by area under curve and the Pearson and Hosmer-Lemeshow tests. Internal validation was performed by bootstrap technique. 4. Predictive score: A predictive score was created from the best-fit model. The coefficient of each predictive variable was divided by the smallest coefficient in the model and allocated a number based on rounding this to the nearest integer. The overall risk score was obtained by summing the integer scores so obtained from all coefficients. 5. The predicted probability of GMG within 2 years for each of the 4 possible overall risk scores was calculated from the predicted odds, that is, probability = (odds)/(1 + odds). Using a receiving operator characteristic (ROC) analysis, the cut point of the overall risk score for the optimal combination of sensitivity and specificity was determined. From this, 2 risk groups are created. This study was approved by the regional ethics committee and by the respective institutional research committees. RESULTS Two hundred sixty patient records were reviewed; of which, 101 fulfilled the study entry criteria (Fig. 1). All patients had full neurological and ophthalmological examinations, including for subtle features of subclinical generalization. Seventy-eight (77%) patients had at least 1 abnormal investigation (serology, EMG, or edrophonium test) to support the clinical diagnosis of OMG. Twenty-three (23%) patients were diagnosed based on the clinical presentation in keeping with ocular myasthenia (fatigability and variability and weakness of more than 1 extraocular muscle, levator, or orbicularis oculi). These latter patients are classified as "probable MG" except for one who was confirmed as "definite MG" because of the later development of GMG (confirmed on symptoms and clinical examination). The patients' demographics and presenting clinical features of OMG are summarized in Supplemental Digital Content 2, Table E2, http://links.lww.com/WNO/A189. Fifty-seven (56%) of patients were male. The only clinical feature that was significantly different in males and females was the age of onset. The mean age of onset of ocular symptoms was earlier in female patients, at 48 ± 17.1 years, compared with 55 ± 17.0 years in male patients (2-sample t test, P = 0.04); this age of onset was approximately normally distributed for both males and females. The proportion of female patients with autoimmune comorbidities (14/44, 32%) was greater than that in males (10/57, 18%), but this did not reach statistical significance (Fisher Exact test, P = 0.11). 254 Duration from onset of ocular symptoms to first presentation for medical attention was median 19.6 weeks (range: 2 days-13 years). The median follow-up duration from ocular symptom onset was 8.4 years (range: 2-42 years). At last review, 31 (31%) patients had developed GMG at a median interval of 1.31 years (range: 3.5 months-20.2 years) from symptom onset; 19 of these 31 patients (61%) experienced GMG within 2 years of symptom onset. Results of univariable analyses of potential predictive factors identified a priori are presented in Supplemental Digital Content 3, Table E3, http://links.lww.com/WNO/A190. Three factors reached a 10% level of statistical significance: the presence of 1 or more comorbidities (odds ratio (OR), 8.84; 95% CI, 1.12-69.72; P = 0.006); seropositivity (to anti-AChR or anti-MuSK) was 7.24 (95% CI, 1.91-27.41) for abnormal (positive) and 3.75 (95% CI, 0.53-26.12) for equivocal (mildly positive) results, respectively (P = 0.004); thymic hyperplasia was weakly associated with GMG (OR, 7.76; 95% CI, 0.66-90.79; P = 0.09). Based on a priori hypothesis, a multivariable logistic regression model was created by first inputting potential predictors with statistical significance at 10%. Variables that would lead to collinearity were not included. The logistic model with 3 predictors, seropositivity, presence of comorbidities, and thymic hyperplasia, showed a model x2, P = 0.001; area under ROC curve of 0.74 (95% CI, 0.64-0.84); Hosmer-Lemeshow goodness of fit, P = 0.90. Owing to missing data on thymic hyperplasia in 15 patients (15%), MIs were performed and the MI model (model x2, P = 0.01; See Supplemental Digital Content 4, Table E4, http://links.lww.com/WNO/A191) had good agreement with the nonimputed model. Because there was little difference between the coefficients for the original and the multiple-imputed model, the following analyses of possible interactions, confounders, goodness of fit tests, and bootstrap analyses were performed on the original model. This model with 3 covariates was compared with more complex models adjusted for age of onset, gender, and ethnicity and tested for interactions with these factors. Likelihood ratio tests showed that the more complex model with interactions or adjustments for age of onset, sex, and ethnicity was not significantly different from the simpler model; this simpler model without interaction and adjustments was therefore selected to avoid overfitting of the model (15). This final logistic model was internally validated by bootstrapping (16). The model from bootstrap tests showed concordance with the logistic model, that is, a good overall predictive model (model x2, P , 0.001); bootstrapping tests also confirmed the importance of all 3 predictors in the logistic model, that is, significant Wald test in the bootstrap model for each of the 3 predictors: any comorbidities, P = 0.006; seropositivity, P = 0.02; thymic hyperplasia, P = 0.003. The area under the ROC curve for the bootstrap Wong et al: J Neuro-Ophthalmol 2016; 36: 252-258 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution FIG. 1. Flowchart of recruitment and outcomes of patients with ocular myasthenia gravis. model was 0.74 (95% CI, 0.64-0.85), and the coefficients and their standard errors for the bootstrap model did not change from the predictive model. All these passed the internal validation tests with regard to calibration and discrimination (16). We used the coefficients from the MI model for creating the risk score. A risk score was calculated for each predictive variable by dividing the estimated coefficient of each variable in the logistic model by the lowest coefficient, that is, 1.69 (corresponding to the coefficient for the presence of thymic hyperplasia) and rounding the quotient to the nearest integer. These individual scores were summed to give an overall risk score for the patient. Table 1 shows the predicted probability for GMG within 2 years corresponding to each of the 4 possible overall risk scores (i.e., total scores of 0, 1, 2, and 3). The 2 risk groups created from an ROC analysis were "high risk" (total scores 2 and 3) and TABLE 1. Predicted probability of GMG within 2 years according to total ROG score Total Score 0 1 2 3 N Total GMG Cases (%) Predicted Probability SD 13 33 38 2 0 (0) 4 (12) 13 (34) 2 (100) 0.018 0.094 0.370 0.780 0.0 0.0005 0.0065 0 ROC area, 0.74 (95% CI, 0.64-0.84); ROC cut point at predicted probability .0.3688 showed a 79% sensitivity and 63% specificity. GMG, generalized myasthenia gravis; ROC, receiving operator characteristic; ROG, risk of generalization; SD, standard deviation. Wong et al: J Neuro-Ophthalmol 2016; 36: 252-258 "low risk" (total scores 0 and 1). The predicted probability for GMG within 2 years for "high risk" was 0.39 ± 0.09 and "low risk" was 0.07 ± 0.03, which corresponds to a negative predictive value of 91% (95% CI, 79-98), a positive predictive value of 38% (95% CI, 23-54), sensitivity of 79% (95% CI, 54-94), specificity of 63% (95% CI, 50-74), and ROC area of 0.74 (95% CI, 0.64-0.85). The difference between the highest predicted probability (0.80 for the group with a total risk score of 3) and the lowest predicted probability (0.02 for the group with a total risk score of zero) was 0.78, that is, a large difference that is considered a good marker of model performance (17). Figure 2 demonstrates good calibration of the risk score, according to the observed outcome and the predicted probability of GMG within 2 years from ocular disease (18). DISCUSSION We present the first prognostic model and predictive score to stratify the risk of GMG in patients with isolated ocular symptoms for the first 3 months of disease. Although our study has limitations and biases, it has shown by proof of principle that stratification of GMG risk can be a useful future direction, for example, in trials on risk-modifying treatments. Prognostic models can contribute usefully to the management of medical conditions, complement clinical reasoning and decision-making, and are increasingly used for selecting suitable patients for early interventional treatment (16,19-23). The association of GMG risk with seropositivity confirms previous reports (9,11). We combined both anti-AChR 255 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution FIG. 2. Calibration of predictive score according to observed outcome and predicted probability of generalized myasthenia gravis within 2 years from onset of ocular disease. The line indicates the predicted probability from the logistic model. and anti-MuSK antibodies because of small sample size of anti-MuSK patients, to improve the statistical strength and association, a technique appropriate for prognostic modeling purposes (13). Future studies with larger numbers may allow separate analysis of anti-MuSK and anti-AChR antibodies. This finding of anti-MuSK antibodies also highlights that pure OMG can occur with anti-MuSK disease (6). Our study also showed that patients with high levels of seropositivity to anti-AChR antibodies were at greater risk of GMG compared with patients with borderline or low seropositivity, again confirming the observations by others (5,24). Ten patients in our cohort had low positive or borderline abnormal anti-AChR antibody result; of which, 4 (40%) developed GMG (2/4 within 2 years). Seven patients had the anti-AChR test repeated; of which, 6 patients had a negative result on either the first or the second test. This highlights the importance of repeating anti-AChR antibody test for initially seronegative patients. Newer serological developments especially with the cell-based assays may further improve the diagnostic accuracy rate in OMG (6). The presence of comorbidities at disease onset has not been previously described as a risk factor for conversion to GMG. We confirmed the significance of this risk factor on the logistic regression (multivariable MI model; OR 6.49; 95% CI, 0.78-53.90; P = 0.08). Adjusting this analysis for age did not affect the statistical significance. The presence of comorbidities at disease onset may be a surrogate marker for the biological tendency for worse outcome. This result was not due to the presence of autoimmunity or age of onset. Analyses did not reveal 256 statistically significant associations with potential confounding factors or interactions with age, gender, ethnicity, and other predictive factors. This result was consistent on both internal validation and survival analysis. This link will need to be further validated externally for more robust confidence. Additionally, we are planning further studies to assess this as an independent risk factor. Thymic hyperplasia was weakly associated with GMG on univariable logistic regression (P = 0.09). The association of thymic pathology with MG has been reported in thymoma, and there is a suggestion that thymectomy of nonthymomatous MG can improve remission rates in MG, although this effect is currently being investigated by a prospective trial (clinicaltrials.gov identifier NCT00294658, accessed January 20, 2015). Recent thymectomy studies have demonstrated a possible disease modifying effect (6,25), further supporting the importance of thymic hyperplasia as a risk factor. None of our patients had thymoma. During study screening, 5 patients with thymoma were detected, but as thymectomy was performed before the onset of GMG or before 2 years, these patients did not fulfill the study inclusion criteria. Although thymic hyperplasia has been associated with seropositivity to antiacetylcholine receptor antibodies (26), we did not find such an association in our patients (P = 0.70 on Fisher Exact test and no collinearity in logistic model). Our patient cohort also allowed observation of the natural history of OMG. Apart from a few (1,4,8,27), most studies of patients with OMG include immunomodulating treatment, which could have altered the course of the disease (5,9,24,28). However, caution should be taken with Wong et al: J Neuro-Ophthalmol 2016; 36: 252-258 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution conclusions from observation of our cohort, as a proportion of patients were excluded because of early treatment with corticosteroids. The proportion of our patients who converted to GMC was similar to studies where a minimum of 3 months of isolated ocular disease was an inclusion criterion (4,5). In our study, 31 (31%) patients developed GMG within their followup period; of which, 19/31 (61%) occurred within 2 years. This was in agreement with findings of Oosterhuis (4) and Sommer et al (5). In studies that did not specify a minimum limit of isolated OMG, a higher rate of GMG occurred: Grob et al (27) described 66% of 202 ocular patients developed GMG; of which, 58% occurred within 6 months, 20% during the remainder of the first year, 7% during the second and third year, 6% beyond the third year. Bever et al (1) had 53/ 142 (38%) patients who developed GMG; of which, 44/53 (83%) occurred within the first 2 years. Our study selection criteria of minimum 3-month isolated ocular disease may therefore account differences in conversion to GMG with some other studies. This 3-month duration is in keeping with the views of other researchers (4,5,29), although the chosen cutoff is necessarily arbitrary as it can be argued that early conversion to GMG should be considered primary generalized disease. From a practical viewpoint, the decision regarding whether to treat with corticosteroids before generalization is less problematic the earlier the generalization occurs. However, this is an important question that should be clarified in future studies. We also included patients who were diagnosed clinically in the presence of fatigability; variability; more than 1 extraocular muscle, levator, or orbicularis oculi weakness, but had negative supportive investigations. This group of patients had a lower rate of conversion to GMG compared with those with at least 1 abnormal supportive investigation (P = 0.066). Such patients may be diagnosed late or perhaps not well represented in larger epidemiological studies, that is, skewing the rate of GMG conversion. It also is possible that there may be some cases in our study with a falsepositive diagnosis. We consider this to be an important group to study, which are not uncommonly problematic in clinical practice and recommend that these data are analyzed separately in future larger prospective studies. The age of onset of ocular symptoms was approximately normally distributed, with a later peak of onset at a mean age of 55 ± 17 years in male and 48 ± 17 years in female patients (P = 0.04), comparable with previous studies (1,27,30,31). Our analyses did not show age as a risk factor, unlike the observation of Bever et al (1) of increased risk after age 50 years but in accord with others (9). This was confirmed by univariable analysis and multivariable analysis (as continuous or binary variable at before and after 50 years of onset). We also did not find a difference in risk after including gender as a covariate. No correlation was found between the time to diagnosis and the age of onset of OMG. Wong et al: J Neuro-Ophthalmol 2016; 36: 252-258 Severe symptoms at onset have been reported as a risk factor for GMG (5). We were unable to assess this because of the nature of data collection, that is, risk of bias if this was not explicitly stated in the notes. This potential risk factor should be explored in future prospective studies. We did not find EMG results to be a predictor for GMG on multivariable survival analysis, although univariable log-rank test hinted at a possible association. An association was found between seropositivity and abnormal jitter on single-fiber EMG (P = 0.02). This association between abnormality on EMG and seropositivity was also noted by Evoli et al (24). An inherent limitation of this study is that not all EMG techniques were standardized, that is, not all patients had repetitive stimulation or singlefiber EMG. The guideline for number of predictors to include in a model is 10% of the outcome of interest in each of the response categories, that is, in this case, 10% of 19 patients who developed GMG within 2 years (for the logistic model) or 30 patients for the survival analyses (16,32). To not lose information of interest too early in the data analysis, we judged it reasonable to keep 3 predictors for the logistic model, instead of 2, according to the above guidelines, because of the statistical significance on univariate analysis for the third predictor of thymic hyperplasia. The eventual "high-risk" category was not affected by the presence or absence of thymic hyperplasia (See Supplemental Digital Content 4, Table E4, http://links.lww.com/WNO/A191). This study enabled us to calculate the sample size needed for a RCT on a risk-modifying treatment to prevent GMG within 2 years. Optimal sample size was calculated on the basis of a Chi-squared test at 5% level of significance and 80% power, for each of the 3 factors: 1) seropositivity; 2) presence of 1 or more comorbidities; and 3) thymic hyperplasia, and considered a treatment to be successful if it shows a 50% reduction of GMG within 2 years, compared to the numbers that converted to GMG in this untreated cohort. This resulted in a sample size of 242, 304, and 68, respectively. Therefore, the minimum number needed for an RCT would be 304 patients (152 treated and 152 untreated). STATEMENT OF AUTHORSHIP Category 1: a. Conception and design: S. H. Wong, A. Petrie, and G. T. Plant; b. Acquisition of data: S. H. Wong and G. T. Plant; c. Analysis and interpretation of data: S. H. Wong, A. Petrie, and G. T. Plant. Category 2: a. Drafting the manuscript: S. H. Wong; b. Revising it for intellectual content: S. H. Wong, A. Petrie, and G. T. Plant. Category 3: a. Final approval of the completed manuscript: S. H. Wong, A. Petrie, and G. T. Plant. All authors contributed to the design of the study. S. H. Wong. performed the retrospective review of patient records and statistical analysis with supervision from A. Petrie. S. H. Wong wrote the first draft of the manuscript, which was revised and reviewed by A. Petrie. and G. T. Plant. 257 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited. Original Contribution ACKNOWLEDGMENTS S. H. Wong previously received a bursary from the Myasthenia Gravis Association (Myaware). This study was presented at the North American Neuro-ophthalmology Society 2015 annual meeting. REFERENCES 1. Bever CT, Aquino AV, Penn AS, Lovelace RE, Rowland LP. Prognosis of ocular myasthenia. Ann Neurol. 1983;14:516-519. 2. Ferguson FR, Hutchinson EC, Liversedge LA. Myasthenia gravis; results of medical management. Lancet. 1955;269:636-639. 3. Grob D, Brunner N, Namba T, Pagala M. Lifetime course of myasthenia gravis. Muscle Nerve. 2008;37:141-149. 4. Oosterhuis HJ. The natural course of myasthenia gravis: a long term follow up study. J Neurol Neurosurg Psychiatr. 1989;52:1121-1127. 5. Sommer N, Sigg B, Melms A, Weller M, Schepelmann K, Herzav V, Dichgans J. Ocular myasthenia gravis: response to long-term immunosuppressive treatment. J Neurol Neurosurg Psychiatr. 1997;62:156-162. 6. Wong SH, Huda S, Vincent A, Plant GT. Ocular myasthenia gravis: controversies and updates. Curr Neurol Neurosci Rep. 2014;14:421. 7. Kerty E, Elsais A, Argov Z, Evoli A, Gilhus NE. EFNS/ENS Guidelines for the treatment of ocular myasthenia. Eur J Neurol. 2014;21:687-693. 8. Grob D. Course and management of myasthenia gravis. J Am Med Assoc. 1953;153:529-532. 9. Kupersmith MJ, Latkany R, Homel P. Development of generalized disease at 2 years in patients with ocular myasthenia gravis. Arch Neurol. 2003;60:243-248. 10. Huang X, Liu WB, Men LN, et al. Clinical features of myasthenia gravis in southern China: a retrospective review of 2,154 cases over 22 years. Neurol Sci. 2013;34:911-917. 11. Guptill JT, Sanders DB, Evoli A. Anti-MuSK antibody myasthenia gravis: clinical findings and response to treatment in two large cohorts. Muscle Nerve. 2011;44:36-40. 12. Leite MI, Waters P, Vincent A. Diagnostic use of autoantibodies in myasthenia gravis. Autoimmunity. 2010;43:371-379. 13. Compston DA, Vincent A, Newsom-Davis J, Batchelor JR. Clinical, pathological, HLA antigen and immunological evidence for disease heterogeneity in myasthenia gravis. Brain. 1980;103:579-601. 14. Garlepp MJ, Dawkins RL, Christiansen FT, Lawton J, Luciani G, McLeod J, Bradley J, Genkins G, Teng CS. Autoimmunity in ocular and generalised myasthenia gravis. J Neuroimmunol. 1981;1:325-332. 15. Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361-387. 16. Moons KG, Kengne AP, Woodward M, Roystan P, Vergouwer Y, Altman DG, Grobbee DDE. Risk prediction models: I. 258 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. Development, internal validation, and assessing the incremental value of a new (bio)marker. Heart. 2012;98:683-690. Altman DG, Royston P. What do we mean by validating a prognostic model? Stat Med. 2000;19:453-473. Tripepi G, Jager KJ, Dekker FW, Zoccali C. Statistical methods for the assessment of prognostic biomarkers(part II): calibration and re-classification. Nephrol Dial Transplant. 2010;25:1402-1405. Moons KG, Royston P, Vergouwe Y, Grobbee DE, Altman DG. Prognosis and prognostic research: what, why, and how? BMJ. 2009;338:b375. Lee KL, Pryor DB, Harrell FE, Califf RM, Behar V, Floyd ML, Morris JJ, Waugh RA, Whalen RE, Rosati RA. Predicting outcome in coronary disease. Statistical models versus expert clinicians. Am J Med. 1986;80:553-560. Rahman M, Simmons RK, Harding AH, Wareham NJ, Griffin SJ. A simple risk score identifies individuals at high risk of developing type 2 diabetes: a prospective cohort study. Fam Pract. 2008;25:191-196. Perel P, Prieto-Merino D, Shakur H, et al. Predicting early death in patients with traumatic bleeding: development and validation of prognostic model. BMJ. 2012;345:e5166. Johnston SC, Rothwell PM, Nguyen-Huynh MN, et al. Validation and refinement of scores to predict very early stroke risk after transient ischaemic attack. Lancet. 2007;369:283-392. Evoli A, Tonali P, Bartoccioni E, Monaco Lo M. Ocular myasthenia: diagnostic and therapeutic problems. Acta Neurol Scand. 1988;77:31-35. Liu Z, Feng H, Yeung SCJ, et al. Extended transsternal thymectomy for the treatment of ocular myasthenia gravis. Ann Thorac Surg. 2011;92:1993-1999. Huang GZ, Lo YL. Correlation between acetylcholine receptor antibody levels and thymic pathology in myasthenia gravis: a review. J Clin Neuromuscul Dis. 2013;14:209-217. Grob D, Arsura EL, Brunner NG, Namba T. The course of myasthenia gravis and therapies affecting outcome. Ann N Y Acad Sci. 1987;505:472-499. Simpson JA, Westenberg MR, Magee HR. Myasthenia gravis: an analysis of 295 cases. Acta Neurol Scand. 1966;42(suppl 23): 7-27. Monsul NT, Patwa HS, Knorr AM, Lesser RL, Goldstein JM. The effect of prednisone on the progression from ocular to generalized myasthenia gravis. J Neurol Sci. 2004;217:131-133. Oosterhuis HJ. Observations of the natural history of myasthenia gravis and the effect of thymectomy. Ann N Y Acad Sci. 1981;377:678-690. Grob D, Brunner NG, Namba T. The natural course of myasthenia gravis and effect of therapeutic measures. Ann N Y Acad Sci. 1981;377:652-669. Bradburn MJ, Clark TG, Love SB, Altman DG. Survival analysis part II: multivariate data analysis-an introduction to concepts and methods. Br J Cancer. 2003;89:431-436. Benatar M. A systematic review of diagnostic studies in myasthenia gravis. Neuromuscul Disord. 2006;16:459-467. Wong et al: J Neuro-Ophthalmol 2016; 36: 252-258 Copyright © North American Neuro-Ophthalmology Society. Unauthorized reproduction of this article is prohibited.