Clinical trials of novel treatments among patients with gMG who were intolerant to or with inadequate response to prior CT were used as data sources for the current analysis. The following trials were included: Phase III trials of eculizumab (REGAIN; ClinicalTrials.gov identifier: NCT01997229) [11]; efgartigimod (ADAPT, NCT03669588; ADAPT-SC, NCT04735432) [12, 20]; ravulizumab (CHAMPION-MG; NCT03920293) [14]; rozanolixizumab (MycarinG, NCT03971422) [16]; and zilucoplan (RAISE, NCT04115293) [18] (Table S1). Eculizumab and efgartigimod PH20 SC were only included in the sensitivity analysis due to differences in study population and/or trial design of the respective pivotal studies. With the exception of ADAPT-SC, where efgartigimod PH20 SC was compared with efgartigimod IV, all the trials were placebo-controlled studies comparing active treatment to placebo, with CT as background therapy in each arm (Table S1). CT could include AChE inhibitors, non-steroidal immunosuppressive drugs (NSAIDs), steroids, or combinations of these treatments.

This study used previously collected or publicly published summary data; thus, no ethical review was required. All the original trials included in this study received institutional review board approval and patient consent. The study was performed in accordance with the Helsinki Declaration of 1964 and its later amendments.

Commonly used primary, secondary, and exploratory endpoint measures in the trials were analyzed whenever feasible, including Myasthenia Gravis-Activities of Daily Living (MG-ADL) and Quantitative Myasthenia Gravis (QMG) scores. MG-ADL is typically used as the primary endpoint in MG trials due to strong regulatory consensus that it should guide trials focusing on disability improvement [21], and QMG is a common secondary endpoint in clinical trials. The MG-ADL is an 8-item patient-reported outcome assessing MG symptoms and their impact on activities of daily living over the prior 7 days [22]. The total MG-ADL score ranges from 0 to 24, with higher scores indicating more disability. The QMG is a quantitative examiner assessment of patient function across 13 domains, based on the strength and endurance of specific muscle groups (i.e., eyes, facial muscles, swallowing, speech, arms, hands, neck, legs, and respiratory) [23]. The total QMG score ranges from 0 to 39, with higher scores indicating more severe disease.

Categorical efficacy outcomes, including the proportions of patients achieving a ≥ 3- and ≥ 5-point improvement in MG-ADL scores, as well as a ≥ 3- and ≥ 5-point improvement in QMG scores, were analyzed in the NMA and used for the estimation of NNT. Results for ≥ 3- and ≥ 5-point improvement in MG-ADL are presented in the main text while the results for ≥ 2-point improvement are presented in the Supplementary Material. Additional efficacy outcomes including continuous measures of change from baseline in MG-ADL, QMG, and Myasthenia Gravis Composite (MGC) scores, and ≥ 2-point improvement in MG-ADL score, were also analyzed in the NMA but not used for NNT estimation.

Because the trials used varying efficacy assessment timepoints (Table S1), the analyses were conducted using efficacy data measured at the primary timepoint of assessment for each trial. For example, Week 4 was selected for efgartigimod to align with the primary assessment timepoint of the ADAPT trial, and to reflect the expected efficacy profile when patients remained on efgartigimod treatment (after Week 4, a high proportion of patients remained off efgartigimod treatment based on the ADAPT trial protocol) [14]. The primary efficacy endpoints were assessed at Week 4 for ADAPT-SC, at Week 6 for MycarinG, at Week 12 for RAISE, and at Week 26 for REGAIN and CHAMPION-MG [11, 16, 18, 20]. This inconsistency on the primary assessment timepoint across trials cannot be addressed in a meta-analysis relying on the existing data. Consequently, a recent NMA of treatment options for gMG also employed the same approach of utilizing the primary timepoint from each trial [21].

Safety outcomes assessed in the NMA for NNH estimation included the proportions of patients with any serious AE (SAE), any treatment-emergent AE (TEAE), and the following specific AEs: headache, nasopharyngitis, nausea, diarrhea, upper respiratory tract infection, and urinary tract infection. In this case, any TEAE was used interchangeably as any AE when extracting from the respective sources. The safety outcomes were selected based on the trial publications and the treatments’ prescribing information, with reported availability in at least three trials including ADAPT.

NMA is one of the most commonly used indirect treatment comparison approaches when head-to-head trials are not available. It synthesizes both direct and indirect evidence from aggregated data, allowing simultaneous comparisons across multiple trials connected within a network. The current NMA assessed risk differences for categorical outcomes and mean differences for continuous outcomes. Both fixed and random effects models were explored, and fixed effects models were selected due to the simple star-shaped configuration of the network with only one link between any two given treatments. For each outcome, the NMA results included the posterior estimates for each treatment within the network with 95% credible intervals (CrIs), pairwise treatment comparison with 95% CrIs, ranking probability to be the best treatment, and the surface under the cumulative ranking (SUCRA) values. The SUCRA value represents the probability that a treatment is among the best options. A SUCRA value of 100% indicates the treatment is certain to be the most effective in the network, while a value of 0% indicates it is certain to be the least effective. The NMA networks are presented in Figure S1.

NNT represents the number of patients who need to be treated in order to achieve one additional improved outcome relative to placebo, and is calculated as the reciprocal of the difference in response rates, informed from the NMA analysis, between each active treatment and placebo [24]. Conversely, NNH refers to the number of patients who need to be treated, relative to placebo, for one additional patient to experience an undesired outcome (e.g., an AE) [24]. It is calculated as the reciprocal of the difference in rates of AE outcomes, informed by the NMA analysis, between each active treatment and placebo. When the AE rate is lower for the active treatment than for placebo, the NNH is a negative value. NNH is not estimable when the AE rates between active treatment and placebo are the same or very similar.

CPIO permits comparison of cost-effectiveness across different treatments, leveraging the concept of NNT. CPIO is calculated as the ratio of the differences in costs and in outcome rates between each active treatment and placebo. For the analysis of CPIO, drug acquisition and administration costs were obtained from IBM Micromedex RED BOOK [25] and the Centers for Medicare & Medicaid Services Physician Fee Schedule [26]. Dosing schedule, treatment duration, and annual costs were based on evidence from the aforementioned clinical trials and supplemented by real-world data [11, 12, 14, 16, 18, 20, 27]. Annual drug and administration costs were considered in the estimation of CPIO (Table S2).

Efficacy and safety outcomes of rozanolixizumab 7 mg/kg and 10 mg/kg dosages were analyzed in the NMA separately reflecting the MycarinG trial design. The rozanolixizumab product label, however, recommends dosing per patient’s body weight and does not necessarily encompass either 7 mg/kg or 10 mg/kg. In addition, NNT and NNH by nature are population-level measures, therefore, the two rozanolixizumab dosing arms were combined for the NNT, NNH, and CPIO analyses to reflect that there were patients with different weights in the population. The combined estimates for rozanolixizumab in NNT, NNH, and CPIO were calculated post-NMA using a weighted average of the 7 mg/kg and 10 mg/kg NMA results.

Incremental NNT is the difference in NNT compared to the best performing treatment, i.e., the treatment with the lowest NNT as the reference arm, for each efficacy endpoint. The incremental CPIO versus efgartigimod IV was also calculated. Efgartigimod IV was selected as the reference arm for CPIO as its CPIO value was lowest across treatments in the primary analysis. The 95% confidence intervals (CIs) of the differences in NNT and CPIO were estimated based on NMA convergence diagnostic and output analysis output with 95% CrIs. Costs were estimated in 2024 US dollars.

Sensitivity analyses were conducted for NNT, NNH, and CPIO, which included efgartigimod PH20 SC and eculizumab in the NMA. Efgartigimod PH20 SC and eculizumab were only included in sensitivity analysis due to differences in their pivotal trials’ designs comparing with the other treatments’ trials. Specifically, the REGAIN trial of eculizumab only enrolled refractory patients with gMG, while the other trials enrolled both refractory and non-refractory patients. Similarly, ADAPT-SC was the only trial without placebo as a comparator and was a non-inferiority trial comparing efgartigimod IV and efgartigimod PH20 SC.

NMAs were conducted using R software v4.3.2 (The R Foundation, Vienna, Austria). The estimations of NNT, NNH, and CPIO were conducted using Microsoft Excel (Microsoft, Redmond, WA, US).