A total of 435 patients (median [IQR] age at onset: 44.1 [29.3, 54.9] years; 56.3% female) were enrolled in the study. The basic characteristics of enrolled patients are shown in Table 1. The percentage of early-onset patients (< 50 years) was 62.8% and of late-onset patients (≥ 50 years) was 37.2%. Compared with male patients, female patients had a younger age of onset, longer disease duration, more concomitant AID and thymic hyperplasia, and more severe MGFA classification (all P < 0.05) (seen in Supplementary Table S1). The median duration of disease from ocular symptom onset to the last follow-up was 7.8 years (IQR: 4.0, 10.7). The positive AChR-ab or RNS was detected in 329 (75.6%) patients and 134 (40.6%) patients, respectively. Thymus status was evaluated in all patients by chest CT scan or thymus pathology, including 126 thymomas, 75 thymus hyperplasia, and 234 normal thymuses. Ptosis was the most common initial presentation in 245 patients (56.3%). More importantly, 237 adult-onset patients with OMG (54.5%) progressed to GMG after a median of 1.1 years (IQR: 0.3, 4.0) (range 0.1–39.1 years). And 31.7% of generalized conversions occurred in the 6 months of onset, 49.8% in the first year, and 65.4% in the first 2 years (Fig. 1A). Among the 237 patients undergoing generalization, 89 (37.6%) patients progressed to type IIA and 96 (40.5%) patients progressed to type IIB according to the MGFA classification. Thymectomy had been performed in 104 patients (23.9%) before the development of generalization and the median (IQR) time from onset to thymectomy was 2.8 months (1.0, 13.0). The maximum clinical severity of patients was classified as ocular MG (MGFA I) in 198 patients (45.5%), mild (MGFA II) in 140 patients (32.2%), and moderate to severe (MGFA III-V) in 97 patients (22.3%). In the last follow-up visit, 263 cases (79.0%) attained MMS or better status, but 77 cases (17.7%) did not improve or worsen.

The Kaplan–Meier curve for time to generalized conversion in adults with ocular myasthenia gravis during the entire study period. A Cumulative rate of generalization free in all patients, the median time of generalized conversion in all patients was 7.5 years. B Age of onset, the median time of generalized conversion in patients aged ≥ 50 years and < 50 was 2.8 and 10.0 years, respectively (P < 0.001). C Gender, the median time of generalized conversion in male and female patients was 10.0 years and 5.0 years, respectively (P = 0.005). D Concomitant autoimmune diseases, the median time of generalized conversion in patients with and without concomitant autoimmune diseases was 2.8 years and 9.0 years, respectively (P < 0.001). E Repetitive nerve stimulation, the median time of generalized conversion in positive and negative patients was 1.9 years and 11.0 years, respectively (P < 0.001). F Anti-acetylcholine receptor antibodies, the median time of generalized conversion in positive and negative patients was 5.0 years and 19.0 years, respectively (P < 0.001). G Thymoma, the median time of generalized conversion in patients with and without thymoma was 1.5 years and 19.0 years, respectively (P < 0.001). H Prednisone therapy, the median time of generalized conversion in patients with and without prednisone therapy was 10.6 years and 3.0 years, respectively (P < 0.001)

The Kaplan–Meier method was used to obtain cumulative probabilities of generalization-free survival (GFS) by stratifying patients according to age at onset (Fig. 1B), gender (Fig. 1C), concomitant AID (Fig. 1D), RNS results (Fig. 1E), AChR-ab status (Fig. 1F), the presence of thymoma (Fig. 1G), and the use of prednisone therapy (Fig. 1H). The analysis demonstrated that patients with early-onset age, concomitant AID, positive AChR-ab or RNS, thymoma, pyridostigmine treatment alone, and female patients had higher conversion rates and earlier time to generalization than those with late-onset age, non-concomitant AID, negative AChR-ab or RNS, non-thymoma, prednisone treatment, and male patients (all P < 0.05). Furthermore, subgroup analysis was performed based on the presence or absence of prednisone treatment, and the results showed that the generalized conversion rate was significantly lower in males, patients without additional AID, and non-thymoma patients in the prednisone-pyridostigmine group than that in the pyridostigmine group (all P < 0.05) (Fig. 2).

Univariable Cox regression analysis and forest plot of potential predictors for generalized conversion in patients with or without prednisone treatment. *P value for interaction of treatment and the factor. There were significant interactions in gender, presence or absence of concomitant AID or thymoma (P < 0.05). Abbreviations: AChR-ab, anti-acetylcholine receptor antibodies; AID, autoimmune disease; CI, confidence interval; HR, hazard ratio; Pre, prednisone; Pyr, pyridostigmine; RNS, repetitive nerve stimulation

The patients were divided into two groups: the OMG-R group (n = 198) and the OMG-G group (n = 237). The common clinical features of the two groups were available in Supplementary Table S2. Age at onset (< 50 years) (HR = 0.631, 95%CI 0.457 ~ 0.782; P = 0.001), male sex (HR = 0.689, 95%CI 0.486 ~ 0.819; P = 0.006), disease duration (HR = 1.018, 95%CI 0.991 ~ 1.045; P = 0.194), concomitant AID (HR = 1.694, 95%CI 1.245 ~ 2.306; P < 0.001), RNS abnormalities (HR = 2.297, 95%CI 1.773 ~ 2.976; P < 0.001), positive AChR-ab (HR = 2.252, 95%CI 1.584 ~ 3.202; P < 0.001), thymoma (HR = 2.931, 95%CI 2.256 ~ 3.809; P < 0.001), prednisone therapy (HR = 0.555, 95%CI 0.429 ~ 0.718; P < 0.001), and thymectomy (HR = 1.635, 95%CI 1.239 ~ 2.159; P < 0.001) were found to be associated with the generalized conversions using univariate cox regression analysis (Table 2). Further multivariate cox regression analysis revealed that concomitant AID (HR = 1.554, 95%CI 1.128 ~ 2.141; P = 0.007), RNS abnormalities (HR = 2.020, 95%CI 1.549 ~ 2.632; P < 0.001), positive AChR-ab (HR = 1.541, 95%CI 1.070 ~ 2.220; P = 0.020), and thymoma (HR = 2.510, 95%CI 1.920 ~ 3.281; P < 0.001) were independent risk factors for the development of generalized myasthenia; while onset age < 50 years (HR = 0.598, 95%CI 0.457 ~ 0.782; P < 0.001), male sex (HR = 0.686, 95%CI 0.523 ~ 0.901; P = 0.007), and prednisone therapy (HR = 0.556, 95%CI 0.427 ~ 0.723; P < 0.001) predicted the reduction of the risk of generalized conversions.

The above 7 clinical predictive factors of generalized conversions were used to form a GFS estimation nomogram (Fig. 3). The performances of this nomogram were assessed by C-index, AUC, and calibration plots. To be specific, the value of the C-index and AUC ranges from 0.5 to 1.0, with 0.5 indicating random chance and 1.0 demonstrating perfect discrimination; the calibration curve showed that the closer each point is to the 45-degree line, the better the consistency between the predicted and actual probability. Our results showed that the C-index of the predictive model was 0.736 (95% CI 0.703 ~ 0.769). Subsequently, we drew the ROC curves of the nomogram for the prediction of 6-, 12- and 24-month GFS with the AUC value indicated (Fig. 4). The AUC for predicting 6-, 12-, and 24-month GFS was 0.765, 0.774, and 0.798, respectively, indicating good discrimination of the predictive model. The calibration plots also showed excellent agreement between the predicted probability of GFS and actual observation, which indicated favorable calibration of the model (Fig. 5). Finally, a dynamic web-based survival rate calculator based on the nomogram was established to predict long-term GFS (https://bizhuajin.shinyapps.io/DynNomapp/). For instance, a 40-year-old male with OMG, who had combined with thymoma and other AID, tested positive for RNS and AChR-ab. The 6-month GFS rate was approximately 0.226 (95% CI 0.118–0.430) if he received early prednisone therapy, whereas the 6-month GFS rate was approximately 0.068 (95% CI 0.021–0.221) if he did not receive early prednisone therapy (Fig. 6).

The nomogram model for predicting generalization in adults with OMG. To use the nomogram, find the position of each variable on the corresponding axis, draw a line to the points axis for the number of points, add the points from all of the variables, and draw a line from the total points axis to determine the GPS probabilities at the lower line of the nomogram. Abbreviations: AChR-ab, anti-acetylcholine receptor antibodies; AID, autoimmune disease; CI, confidence interval; GFR, generalization-free survival; Pre, prednisone; RNS, repetitive nerve stimulation

ROC curve and AUC of the nomogram for the prediction of 6-month GFS (A), 12-month GFS (B), and 24-month GFS (C)

Calibration curves of the nomogram for the prediction of 6-, 12-, and 24-month GFS (A–C). The abscissa represents the nomogram-predicted GFS rate, the ordinate represents the actual GFS rate, and the calibration curves for 1-, 12-, and 24-month GFS rates showed satisfactory agreements between the predicted and actual values

An example to illustrate the use of the web-based GFS rate calculator. A A 40-year-old male with OMG, who had combined with thymoma and other autoimmune diseases, tested positive for RNS and AChR-ab. B His generalization-free survival curve based on whether he received prednisone (a) or not (b). C The 6-month GFS rate was approximately 0.226 (95% CI 0.118–0.430) if he received prednisone treatment, whereas the 6-month GFS rate was approximately 0.068 (95% CI 0.021–0.221) if he did not receive prednisone treatment