Nomogram for short-term outcome assessment in AChR subtype generalized myasthenia gravis

There are 1193 MG patients registered in a tertiary referral diagnostic center in Huashan Hospital from August 8, 2012, through December 18, 2020. The inclusion criteria were (1) onset symptoms and signs compatible with gMG; (2) immunotherapy naive at baseline; (3) seropositive for anti-AChR antibody; (4) MG-ADL score > 1 at baseline; (5) follow-up period longer than half a year from baseline; (6) exclusion of other MG mimicking diseases including Lambert–Eaton myasthenic syndrome, peripheral neuropathy, myopathies, and motor neuron diseases. Eligible patients with the integrated baseline data recruited from February 13, 2017, through August 2, 2019, were included in the training cohort for the development of the nomogram, and those recruited from August 2, 2019, through March 13, 2020, were included into the temporal validation cohort. Then, the nomogram was externally validated using 45 anti-AChR antibody-positive gMG patients who have not received immunotherapy from May 2015 to May 2021 at 4 tertiary centers in China (Xiangya Hospital, Xuanwu Hospital, Tangdu Hospital, and Wuhan No.1 Hospital).

The clinical baseline variables include gender, age at onset, the comorbidities of autoimmune disease, and disease duration. The age at onset of MG is classified into three subgroups including early-onset (10–49 years), late-onset (50–64 years), and elderly-onset (65 years or older) [12]. The concurrent autoimmune diseases identified in our cohort include Graves' disease, Hashimoto's autoimmune thyroiditis, type 1 diabetes mellitus, immune thrombocytopenic purpura, autoimmune hemolytic anemia, and vitiligo [13]. The disease duration is defined as the period from the onset of weakness symptoms of MG to the first visit to our hospital. The MG associated clinical features include Myasthenia Gravis Foundation of America (MGFA) classification, thymoma concurrence, history of thymectomy, MG worsening, anti-AChR antibodies titers, pyridostigmine dosage, manual muscle test (MMT) score, MG-ADL score, and the related subscores (bulbar, respiratory, ocular, and limb score), and quantitative myasthenia gravis (QMG) score and the related subscores (extraocular muscle, bulbar muscle, gross motor, and axial motor score). The presence of thymoma is determined by a computed tomography scan. MG worsening is defined as a substantial exacerbation in muscle weakness and fatigability, or increased medication [14]. The anti-AChR antibodies titer was measured by enzyme-linked immunosorbent assay (ELISA, Euroimmun, Lübeck, Germany) and the cut-off value was 0.50 nmol/L.

We divided the total MG-ADL score into four subscores: (1) Ocular score: double vision and eyelid droop; (2) Bulbar score: talking, chewing, and swallowing activities; (3) Respiratory score: the activity of breathing; (4) Limb score: the ability to brush teeth or comb hair, and arise from a chair. For QMG score, it was divided into 5 subscores: (1) Extraocular muscle score: first three items (double vision on lateral gaze, ptosis, and facial muscles); (2) Bulbar muscle: score fourth and fifth items (swallowing 4 oz. water, and the onset of dysarthria); (3) Gross motor score: sixth, seventh, ninth, and tenth, twelfth, and thirteenth items (arms outstretched, hands grip, and legs outstretched); (4) Axial motor score: eleventh item (head lifted); (5) Respiratory score: eighth item (Vital capacity, % predicted).

Our analysis showed that the continuous demographic characteristics data in this study were not normally distributed. The missing data of thymoma, thymectomy, anti-AChR Abs titer, and pyridostigmine dosage account for less than 10%. These missing data were missed at random and replaced by the average of the observed values. Continuous variables were expressed as medians (quartiles) and compared between groups using the Mann–Whitney U test. Categorical variables were expressed as frequencies (percentages) and were tested using the χ2 test or Fisher exact test. To determine the cut-off values of the continuous variable, we created receiver operating characteristic curves for “MSE” and defined them as the points on the ROC curve where Youden's index reached the highest. The significance of each variable in the training cohort was analyzed using univariate and multivariate logistic regression analyses. Variables showing statistical (P < 0.1) and clinical significance of the univariate analysis were included in the multivariate logistic regression analysis to develop the nomogram to predict whether a patient would achieve MSE.

The variance inflation factors (VIFs) were generated to examine individual predictors for potentially strong contributions to multicollinearity. The discrimination performance of this nomogram was measured by the concordance index (C-index) in the training and validation cohorts. The Hosmer–Lemeshow test was applied to assess the agreement between nomogram predicted and observed probabilities. All analyses were performed using IBM SPSS version 20.0 (SPSS Company, Chicago, IL, USA) and R software (R version 4.0.3, USA).

A total of 1193 MG patients have been initially registered in our referral center-based database. According to the inclusion flowchart, we finally enrolled 120 AChR subtype gMG patients with no immunotherapies at the baseline registry (Fig. 1). Of these gMG patients, 96 and 24 patients were then included in the training and temporal validation cohort splitting by time. Besides, 45 AChR subtype gMG patients from the other 4 centers were enrolled for external validation.

The baseline clinical characteristics and the outcome of MG patients in the training and temporal validation cohort were comparable, except for the differences in the respiratory score (p = 0.035), and respiratory muscle score (p = 0.002) (Table 1). In the development set, 70 patients (72.9%) achieved MSE and the median disease duration was 7 (3.0–30.5) months. At baseline, 34% of patients were diagnosed to have concurrent thymoma and 24% had undergone thymectomy in the development set. The ADL and QMG scores of the development cohort were 5 (4.0–8.0) and 11 (9.0–14.0), respectively.

The clinical characteristics of the external validation and development groups are summarized in Table 2. The frequency of MSE was similar for the development (72.9%) and external validation groups (73.3%), whereas there were some differences between these groups regarding the frequency of thymectomy, disease duration, anti-AChR Abs titer, MMT score, bulbar score, bulbar muscle score, and gross motor score.

MSE status was achieved in 70 (72.9%), 20 (83.3%), and 33 (73.3%) patients in the training, temporal validation, and external validation cohorts at 12 months after baseline recruitment. For the patients who did not achieve MSE, the median ADL scores were 3 (range 3–6), 4.5 (range 3.25–5), and 2.5 (range 2–7) in the training, temporal validation, and external validation cohorts, respectively.

In the training and temporal validation groups in Huashan Hospital, the initial dose and dose-escalating manner of prednisone depended on the physician's decision. The final oral prednisone dose for each patient was at 0.8 mg to 1 mg/kg and azathioprine, tacrolimus, or mycophenolate mofetil as immunosuppressants concurrent with oral prednisone. Three patients had received rescue therapies including immunoglobulin and plasma exchange.

We identified three risk factors significantly associated with MSE including duration, ocular score, and gross motor score (p < 0.1) (Table 3). Considering the clinical significance, we also included the QMG score (p = 0.155) along with these statistically significant variables into the multivariate logistic regression. All these above variables were independently associated with MSE (p < 0.05), with results reported as odds ratio (95% CI), duration ≤ 12 months (3.45 [1.23–10.24]), ocular score ≤ 2 (6.00 [1.82 to 24.58]), QMG score > 13 (11.95 [2.31 to 95.82]), and gross motor score ≤ 9 (10.82 [2.22–69.13]). The VIFs of them were 1.01, 1.25, 1.78, and 1.84 respectively, suggesting that there was no multiple collinearity among the four independent risk factors. We then used these four factors to establish an individualized prediction nomogram, which can calculate the total point for each gMG patient with anti-AChR antibodies and converted it to predicted probabilities of MSE (Fig. 2). This nomogram was then validated in both the temporal validation cohort derived from Huashan Hospital and the external validation cohort.

The ability of this nomogram to differentiate between patients who do or do not achieve MSE is excellent as the C-indexes are 0.810 (95% CI, 0.72–0.90), 0.944 (95% CI 0.83–1.00), and 0.773 (95% CI, 0.63–0.92) in the development, temporal validation, and external validation cohorts, respectively (Fig. 3). Besides, the p-values of the Hosmer–Lemeshow test are 0.98, 0.99, and 0.61 for the development, temporal validation, and external validation sets, which indicates good agreement between nomogram predicted and observed probabilities.