Network connectivity, between-study heterogeneity and timepoint challenges in generalized myasthenia gravis: a feasibility assessment of indirect treatment comparisons

A systematic literature review (SLR) to identify clinical trials reporting on the efficacy and safety of systemic treatments for patients with gMG was conducted. Detailed SLR methods are provided in the Supplementary Materials. The SLR was designed and reported in accordance with the Preferred Reporting Items for Systematic Literature Reviews and Meta-Analysis (PRISMA) guidelines [38]. An SLR protocol and an ITC protocol were developed a priori. IPD-only analyses were assumed to be infeasible because IPD from RCTs for more than one treatment are seldom available to the same investigators. Hence, ITC feasibility assessment was restricted to NMAs and MAICs/STCs.

We searched biomedical databases, recent neurology conference proceedings (i.e., American Academy of Neurology, American Association of Neuromuscular and Electrodiagnostic Medicine, and European Academy of Neurology from 2021 to 2023), regulatory and HTA agency websites, and ClinicalTrials.gov. Additionally, bibliographies of relevant SLR and ITC publications were checked for relevant trials.

Two independent reviewers screened citations according to prespecified eligibility (PICOS) criteria. The eligible population was patients diagnosed with gMG. Interventions of interest were SOC therapies, add-on immunotherapies and thymectomy. Outcomes of interest included the efficacy outcomes Myasthenia Gravis Activities of Daily Living (MG-ADL) and Quantitative Myasthenia Gravis (QMG). Only clinical trials published in English from 2003 onward (i.e., in the last 20 years) were included as older trials may have limited relevance to current management of gMG.

In alignment with EU guidance, only RCTs were considered for the ITC feasibility assessment as they are the preferred data source for ITCs [21]. The evidence base for the feasibility assessment was comprised of publications of phase II or III RCTs studying US FDA and EMA approved dosing regimens (and routes of administration) of newer targeted immunotherapies added to SOC for gMG, off-label add-on immunotherapies to SOC recommended in published national or international gMG treatment guidelines (i.e., rituximab and non-acute IVIg) and nipocalimab (only the Vivacity-MG3 phase III trial was included) [8,9]. Although results for the nipocalimab Vivacity-MG3 phase III RCT were not published at the time of the SLR searches, this trial was included because a publication became available during preparation of this manuscript [12]. A phase II RCT by Wolfe et al. (2002) was also considered for the feasibility assessment [39]. Although this trial was outside the SLR date limit, it was identified as the only RCT reporting MG-ADL change from baseline (CFB) data for non-acute IVIg and had been included in a previous ITC [22].

Data extraction was performed by a single reviewer and independently assessed for accuracy and completeness by a second reviewer. Extracted data included study, patient and outcome characteristics.

The following trial characteristics were assessed to inform the feasibility of ITCs: study design characteristics, treatment characteristics, trial eligibility criteria, patient characteristics at baseline and outcome characteristics. Heterogeneity in baseline risk (i.e., placebo arm outcomes) was also assessed. Outcomes of interest were MG-ADL and QMG CFB and response (i.e., proportion reaching a within-person meaningful change threshold) as these were the most commonly reported primary or key secondary efficacy outcomes in the RCTs.

We conducted the ITC feasibility assessment in three steps: investigation of reported trial information to identify substantial within- and cross-trial differences; assessment of network connectivity to understand whether anchored ITC methods such as NMA were appropriate; and synthesis of findings and identification of appropriate ITC methods. Our approach was based on guidance from Cochrane, the EU, NICE and the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) [21,28–31,37,40–42].

We first qualitatively assessed whether clinically important differences in patient characteristics at baseline (e.g., age, sex, baseline MG-ADL score, baseline QMG score, prior corticosteroid exposure, prior NSIST exposure) were present across treatment groups within each trial. Potential imbalances were identified by calculating the standardized mean difference (SMD); an SMD > 0.2 was considered a substantial difference [43] and SMDs between 0.1 and 0.2 were also noted. We also investigated whether any patient characteristics were adjusted for in the models used to calculate MG-ADL and QMG outcomes in each trial, as this would help mitigate potential bias due to within-trial imbalances. Next, we assessed between-trial differences in study design characteristics (including treatment details), population characteristics (based on trial eligibility criteria and summary measures of baseline characteristics), and outcome characteristics (definitions and assessment timepoints). We examined reported subgroup analyses for six registration phase III RCTs to identify potential TEMs. This approach helped contextualize the potential importance of any cross-trial differences in baseline patient characteristics. A summary of reported subgroup analyses and the criterion used to identify potential TEMs is provided in Supplementary Table 1. Treatment schedules, timepoints at which CFB outcomes for MG-ADL and QMG were assessed relative to baseline, and the minimum point improvements (MPIs) (i.e., within-person meaningful change thresholds) used to define MG-ADL and QMG responders, were compared across included RCTs.

Network connectivity was assessed by examining overlap in the treatment arms of included RCTs. Network diagrams were used to visualize the connectivity of treatments. The availability of a connected network and its structure are important determinants toward the selection of ITC methodology [30,31]. A generalized network diagram was constructed based on included RCTs independent of outcome availability and potential issues with the validity of common comparators. The characteristics of the placebo arms of the different trials, such as route of administration, as well as placebo response for MG-ADL and QMG outcomes were compared with identify any cross-trial differences that could introduce bias in an anchored ITC.

Findings from these assessments were synthesized and considered along with trial data availability (i.e., aggregate data vs IPD) to identify: appropriate ITC methods; a base case analysis for each method and potential sensitivity analyses; and limitations of each method.

Detailed SLR results are available in the Supplementary Materials. In total, 186 records representing 47 unique trials met the SLR inclusion criteria. Of the 47 trials, 32 were RCTs, seven were open-label extensions of included RCTs and eight were single-arm trials. Fifteen RCTs (including Vivacity-MG3 and Wolfe et al. [2002]) studied add-on immunotherapies and were included in the ITC feasibility assessment. A description of the 19 RCTs captured via the SLR but excluded from the feasibility assessment is provided in the Supplementary Materials. A list of the 15 RCTs included in the feasibility assessment is provided in Supplementary Table 2. A summary of the included trials and their key characteristics is provided in Table 1.

Within-trial imbalances in baseline patient characteristics were noted for several of the included trials (Figure 1). In general, more extensive imbalances (i.e., SMD > 0.2) were observed for smaller trials (e.g., phase II and non-registration trials such as RINOMAX) compared with larger registration phase III trials. For example, six of the nine baseline characteristics we evaluated for the AGRX-113-1602 phase II efgartigimod RCT were imbalanced based on having an SMD > 0.2, whereas only one of the evaluated characteristics was imbalanced in the ADAPT phase III efgartigimod RCT. Although adjustments for baseline MG-ADL and QMG scores were performed in most included trials, other patient characteristics were generally not adjusted for.

A summary of key trial characteristics is provided in Table 1, patient characteristics at baseline in Table 2 and MG-ADL and QMG assessment details in Table 3. Summaries of general study design characteristics, trial eligibility criteria and outcome characteristics for the 15 included RCTs are available in the Supplementary Materials.

Of the 15 RCTs, seven were phase ii trials and eight were phase III trials. Of the phase III trials, ADAPT (efgartigimod), CHAMPION-MG (ravulizumab), MycarinG (rozanolixizumab), RAISE (zilucoplan), REGAIN (eculizumab) and Vivacity-MG3 (nipocalimab) were registration trials. Whereas rozanolixizumab and zilucoplan were subcutaneous interventions, all others were administered intravenously. Dosing schedules and routes of administration varied across the included trials (Figure 2). Notably, in ADAPT, patients received four weekly infusions of efgartigimod per cycle, repeated based on waning MG-ADL response no sooner than 8 weeks after the start of the preceding cycle. In MycarinG, patients completed a 6-week treatment period with rozanolixizumab followed by an 8-week observation period, with outcomes measured throughout. The unique approaches employed in ADAPT and MycarinG pose challenges for conducting ITCs because patients in these trials improved while on treatment but worsened off-treatment, whereas in the trials that used maintenance dosing, patients experienced sustained improvement.

The frequency of administration (maintenance dosing) also varied across trials, ranging from daily in the RAISE zilucoplan trial to every 8 weeks in the CHAMPION-MG ravulizumab trial, with most treatments being administered weekly or every 2 weeks (Figure 2).

All included RCTs reported CFB outcomes for MG-ADL and QMG. The principal timepoints at which these outcomes were assessed varied substantially across trials (Table 3), reflecting in part differences in study duration and dosing schedules. For example, the principal assessment timepoint for these outcomes ranged from 4 weeks in the ADAPT efgartigimod trial up to 52 weeks in the BeatMG rituximab trial.

The principal MPI values defining MG-ADL and QMG response also varied across trials (Table 3). Specifically, phase III trials for C5 complement inhibitors (eculizumab, ravulizumab and zilucoplan) used an MPI of 3 for MG-ADL and 5 for QMG, whereas other trials used MPIs of 2 and 3, respectively.

Trial eligibility criteria (e.g., Myasthenia Gravis Foundation of America [MGFA] class, antibody status, baseline MG-ADL score and treatment history) and patient characteristics at baseline (e.g., age, sex, race/ethnicity, geographical region, MGFA class, antibody status, disease duration, prestudy treatment exposure and disease severity) were generally well-reported across RCTs (Table 2 & Supplementary Table 3). Although assessed patient characteristics were broadly similar across trials, there were key differences in antibody status, time since gMG onset and prior treatment exposure.

Mean age at baseline and the proportion of female patients were reported for all included RCTs and were broadly comparable, with the exception of the RINOMAX rituximab trial population being relatively older and having a lower proportion of females (Table 2). Whereas trials of C5 complement inhibitors and non-acute IVIg consistently included only anti-AChR antibody-positive (AChR+) patients, trials for other treatments were less restrictive (Supplementary Table 3). The registration trials for FcRn inhibitors (ADAPT, MycarinG and Vivacity-MG3) were a mix of AChR+ and anti-AChR antibody-negative (AChR-) patients. However, where reported, data for the AChR+ subpopulation of these trials could be used to better align with other trials in ITC analyses.

The proportion of white patients, the proportion by geographical region, mean duration of disease, MGFA class at baseline, baseline MG-ADL score, baseline QMG score, the proportion receiving corticosteroids at baseline and the proportion receiving an NSIST at baseline were less well reported but broadly comparable across trials, with some exceptions (Table 2). Unlike other included trials, RINOMAX only included patients with new-onset gMG, as reflected by a lower baseline QMG score cutoff for inclusion, shorter average disease duration and lower baseline MG-ADL and QMG scores compared with other trials (Tables 1 & 2). Baseline MG-ADL and QMG scores were also relatively low for the BeatMG rituximab trial and the Bril et al. (2023) and Wolfe et al. (2002) non-acute IVIg trials. Unlike other included trials, the proportion receiving an NSIST at baseline was 0% in RINOMAX and the proportion receiving corticosteroids at baseline was 100% in BeatMG and the Bril et al. (2023) non-acute IVIg trial.

Cross-trial differences in treatment exposure were also noted. RINOMAX was the only trial to not permit prior thymectomy. The REGAIN eculizumab trial included relatively more treatment-refractory patients than the other trials (Table 1) wherein patients were required to have previously failed two or more immunosuppressive therapies or failed one or more immunosuppressive therapies with IVIg or plasma exchange given at least four-times per year for 12 months without symptom control.

Permitted and prohibited background treatments were broadly similar across included trials (Supplementary Table 4). Specifically, all trials permitted corticosteroids and NSISTs. However, the proportion of patients receiving specific background treatments, and their doses, routes and frequencies of administration, were not well reported across trials. Hence, while differences in background treatments may introduce bias in ITCs, it is difficult to assess the extent of the issue.

Prior to consideration of the comparability of placebo arms across included trials, a generalized connected network of treatments was visualized with placebo as a central node (Supplementary Figure 1). The network was star-shaped as all treatments (apart from the subcutaneous form of efgartigimod compared with intravenous efgartigimod in the ADAPT-SC trial) have been compared only with placebo. The network was also found to be sparse, as only one or two trials were available for each pair of connected treatments. Further, the timepoints at which MG-ADL and QMG outcomes were measured and the responder thresholds used differed across included trials (Table 3). These differences in outcomes are expected to impact the connectivity of networks.

While construction of a connected network anchored through placebo appears feasible, the comparability of placebo arms across trials is an important consideration in the conduct of anchored ITCs. For MG-ADL and QMG CFB, placebo response varied substantially across the included registration phase III RCTs (Supplementary Table 5). MG-ADL CFB for the placebo group ranged from -0.78 in the MycarinG rozanolixizumab trial (at week 6) to -3.25 in the nipocalimab Vivacity-MG3 trial (average of weeks 22, 23 and 24). Similarly, QMG CFB for the placebo group ranged from -0.8 in the CHAMPION-MG ravulizumab trial (at week 26) to -3.25 in the RAISE zilucoplan trial (at week 12). The differences in placebo response did not appear to be related to differences in reported trial characteristics, including dosing schedule or route of administration. For example, one might expect intravenous rather than subcutaneous dosing to be associated with a higher placebo response, but this was not consistently the case here. This suggests potential confounding by other sources of cross-trial heterogeneity, such as unmeasured or unreported patient baseline characteristics.

The assessments of trial characteristics and network connectivity identified several important considerations for selecting the most appropriate ITC analysis methods. There was evidence of within-trial imbalances in patient characteristics at baseline across included trials that could potentially bias NMA results. Although imbalances were present across all trials, they were more frequent in smaller non-registration trials. This finding suggests it may be appropriate to only consider large, well-balanced phase III registration trials in NMA base case analyses, while including the other trials in a sensitivity analysis to understand their impact on results. Still, the potential for residual confounding due to the observed imbalances in registration trials remains.

For included phase III registration RCTs, subgroup analyses of MG-ADL and QMG indicated the following characteristics may be TEMs: age, sex, race, geographical region, antibody status, disease duration, baseline MGFA class, baseline MG-ADL score and prior immunosuppressant therapy use (Supplementary Table 1). Several of these characteristics were noted to differ appreciably across included RCTs, which could introduce bias in ITCs due to violation of the exchangeability assumption. For example, the distribution of patients across relevant MG subgroups varied across trials, with some trials only including AChR+ patients. For trials including both AChR+ and AChR- patients, it may be appropriate to match the proportions of these groups in MAICs/STCs. In NMAs, it may be appropriate to restrict base case analyses to the AChR+ subpopulation to ensure comparability with trials of only AChR+ patients, with sensitivity analyses that include full trial populations to understand the impact of heterogeneity in serotypes across trials. As another example, the RINOMAX and REGAIN trial populations differed substantially from other trials with respect to potential TEMs (i.e., age, sex and baseline MG-ADL score for RINOMAX, and prior immunosuppressant therapy use for REGAIN). It may be appropriate to exclude RINOMAX and REGAIN as outlier trials from base case ITCs, with sensitivity analyses in which they are included.

An additional challenge in conducting anchored comparisons such as NMA is related to cross-trial differences in placebo arms (e.g., background treatments, administration route and schedule). Differences in placebo arm characteristics (as well as heterogeneity in measured and unmeasured TEMs) can result in systematic differences in placebo response, which can lead to bias in anchored ITC results. Indeed, substantial differences across the included trials in placebo response rates for MG-ADL and QMG were observed. Since the evidence network for gMG treatments is ‘star’ shaped with all RCTs but one linked via placebo, the validity of NMA is likely to be particularly dependent on the assumption that placebo arms across trials are comparable.

Given the within- and cross-trial variations in baseline characteristics, differences in the implementation of placebo arms, and the observed heterogeneity in placebo response rates across trials, there is a significant risk of bias in the results of anchored ITC approaches such as NMA. Although extensions of NMA such as baseline risk-adjusted NMA and network meta-regression can adjust for differences in TEMs across trials, they are not particularly useful and may not even be feasible in this situation given the sparseness of the evidence network (i.e., only one or two trials for each intervention). For the same reason, exclusion of outlier trials expected to bias ITC results will prevent some treatments (e.g., eculizumab) from being included in NMA, thereby potentially reducing the clinical relevance of ITCs.

Anchored MAICs/STCs, like NMAs, take advantage of the availability of a shared comparator between RCTs to compare relative outcomes in two treatment arms, thus maintaining randomization and the presumed balance of prognostic factors between trials. Unlike an NMA, an MAIC or STC leverages IPD to adjust for differences in reported patient characteristics (i.e., TEMs) that could otherwise bias ITC results. These methods were deemed feasible and are less likely to bias results compared with NMA. However, potential bias due to systematic cross-trial differences in placebo response discussed earlier, cannot be addressed by anchored MAICs/STCs. Similar to NMA, anchored MAICs may also give biased results in the presence of within-trial imbalances in patient characteristics. Arm-level (rather than trial-level) adjustments are a potential approach to address this concern, however they may break randomization, potentially introducing bias by distorting the balance between treatment arms of covariates not subjected to adjustment [52].

As an alternative, unanchored MAICs/STCs can help address the limitations associated with cross-trial placebo arm heterogeneity and imbalances within trials because they do not rely on the use of a shared comparator. Notably, unanchored analyses compare absolute outcomes of treatment arms from different trials, therefore differences in all prognostic factors and TEMs need to be adjusted for to avoid bias in the results [42]. Thus, unanchored adjusted ITC methods may be considered for treatment comparisons in gMG given the identified differences in placebo arm characteristics and within-trial imbalances.

Given the unique strengths and limitations of each ITC method in addressing the challenges posed by gMG trials, the use of multiple methods is warranted to provide a more holistic approach in assessing the comparative effectiveness of gMG treatments (Table 4, Figure 3). Conducting different ITC methods and comparing their results can help to clarify the extent to which methods may be biased. If all methods produce results in the same direction and of similar magnitude, this can increase confidence in the validity and reliability of comparative efficacy estimates. Alternatively, if there are differences in the results from each method, this can suggest further investigation is warranted to identify the most reliable method for the comparison(s) of interest.

Differences in reported timepoints may impact the validity of ITC analyses. For any outcome measured over time, the choice of timepoint is critical to minimize bias from this factor in an ITC. Ideally the same timepoint should be used across trials, but this is difficult in practice given variation in follow-up times and the frequency of outcome assessment across RCTs. Even in situations where outcomes are reported for a comparable timepoint, fluctuations in treatment response due to cyclical dosing schedules may favour one treatment over another in an ITC. The timepoint(s) selected for analysis may reflect a period where treatment is or is not being received. For example, in the ADAPT trial, the primary assessment timepoint (week 4) reflected the end of the treatment cycle but does not account for potential waning of effect during off-treatment. Hence, regardless of the ITC method used, it may be valuable to conduct analyses using outcome data representative of multiple timepoints (e.g., taking the average across several timepoint or using an area under the curve [AUC] approach to compare the longer term efficacy of cyclical vs continuous treatments), as opposed to a single timepoint.

Finally, MG-ADL and QMG response definitions and assessment timepoints for the principal reported outcomes generally varied between RCTs. In particular, trials for C5 complement inhibitors used higher responder thresholds than trials for FcRn inhibitors. It is important to account for this difference when conducting ITCs, which could involve using other reported response data to achieve alignment in MPI where possible. Many of the included trials reported response outcomes for multiple MPI values. However, in cases where it is not possible to align on MPI values, ITCs for response outcomes with varying MPIs may still warrant investigation, so long as the limitations of such an approach are acknowledged.

In this study we performed a formal feasibility assessment to systematically evaluate currently available trials for gMG treatments with the goal of identifying the advantages and disadvantages of different ITC methods. The current study highlights the importance of conducting a thorough ITC feasibility assessment, as failing to do so may lead to use of ITC methods yielding biased results. Several barriers to conducting robust ITCs in gMG were identified, including within-trial imbalances in patient characteristics, small trial sizes and cross-trial differences in potential TEMs. Further, heterogeneity in placebo administration characteristics, background therapies and cross-trial variation in placebo response for key outcomes were noted. Treatment strategies (i.e., cyclical vs continuous), dosing schedules and outcome assessment timepoints were inconsistent across trials, necessitating careful consideration of methods and timepoints when interpreting outcomes. The use of multiple ITC methods, with careful evaluation of underlying assumptions and limitations, can help limit bias and provide a more complete understanding of the comparative effectiveness of gMG treatments.

We also identified the need for methodological approaches that address the heterogeneity in treatment strategy and outcome assessment timepoints across RCTs for gMG therapies. The issues of cyclical versus continuous treatment approaches and varying assessment timepoints are independent of the choice of ITC method (NMA, anchored/unanchored MAIC/STC) as none of these methods can adjust for such differences. As gMG is a chronic disease, timepoints reflecting longer term efficacy are generally of greatest interest to decision makers. However, assessment of long-term efficacy for cyclically-dosed treatments is challenging given that the reported data from trials reflect timepoints associated with maximal effects and/or waning of treatment benefit. The potential for dosing regimens of cyclical therapies in clinical practice to differ from dosing in trials should also be considered. Approaches employing average treatment effects or AUC may hold promise for analyzing data for cyclical therapies in ITCs; however, they do not capture patient or clinician preference for consistent versus fluctuating disease control.

Several of the issues associated with conducting ITCs in gMG identified in our study have been reported in previous HTA agency reviews. Between-trial population differences and their implications for appropriate ITC methods identified in the current study were highlighted by Canada’s Drug Agency (CDA) and NICE in their reviews of efgartigimod and ravulizumab. The CDA efgartigimod review noted results of an NMA comparing efgartigimod, ravulizumab and non-acute IVIg were uncertain due to important differences between gMG trials (including variability in trial eligibility criteria) as well as inconsistent and insufficiently detailed reporting of baseline patient characteristics [22]. The ravulizumab submissions to NICE and CDA included Bucher ITC, MAIC and IPD-only analyses comparing ravulizumab with eculizumab [23,26]. Both HTA agencies noted differences in inclusion criteria related to prior therapy between REGAIN and CHAMPION-MG (i.e., REGAIN required patients to have failed on immunosuppressive drugs, while CHAMPION-MG did not). The critical appraisal by CDA mentioned a small effective sample size for submitted MAIC and IPD-only analyses as evidence of limited overlap between trial populations and concluded an unadjusted ITC between CHAMPION-MG and REGAIN was unlikely to be valid. In reviews of efgartigimod and ravulizumab, NICE noted the availability of two rituximab trials (BeatMG and RINOMAX) but did not push for their inclusion in ITCs due to their small sizes and lack of statistically significant results versus placebo [26,54]. Taken together, the NICE and CDA reviews supported our identification of REGAIN and RINOMAX as potential outlier trials to consider excluding from base case analyses and more generally the importance of conducting adjusted ITCs to account for population differences where possible.

The CDA and NICE ravulizumab reviews also identified the importance of clear and detailed reporting of ITC analyses, including justifications for the ITC methods selected, how trials and outcomes to be compared are identified, and the selection process by which covariates to be included in adjusted analyses are identified [23,26]. Specifically, NICE raised concerns about a less-than-comprehensive set of covariates having been used in adjusted ITC analyses, as well a lack of sensitivity analyses to evaluate the robustness of analyses with adjustments for different sets of covariates. A comprehensive and transparent approach to identifying and justifying covariates for adjustment in MAICs/STCs is critical to optimize the adjustment process and thereby minimize bias introduced by cross-trial differences in patient populations. In a review of published approaches to selecting effect modifiers for adjustment in ITCs, Freitag et al. recommended a three-pronged approach for identifying TEMs: systematic reviews; expert input and assessment of the distribution of TEMs (e.g., examining subgroup analyses in clinical trials) [55].

Aligned with our findings, challenges for ITC analyses due to cross-trial variation in dosing schedules, follow-up times and primary assessment timepoints for MG-ADL and QMG outcomes were noted by both CDA and NICE in their reviews of gMG treatments, as well as in an EMA review of zilucoplan [22–24,26,54]. Specifically, the CDA review of efgartigimod highlighted its unique dosing schedule (i.e., symptom-based cyclical dosing, as opposed to continuous dosing for comparators) and the range of study follow-up times as important differences between trials for gMG treatments that increase uncertainty and limit the interpretability of ITC results [22]. We also noted using the primary assessment timepoint from each trial in an ITC could introduce bias for or against some treatments, whereas conducting ITCs at an early timepoint (e.g., week 4) could bias against treatments demonstrating improved response over time. In their ITC study, Fawsitt et al. discussed the importance of considering different analytical approaches where there are cross-study differences in the timepoints for which CFB outcome data are available, and specifically used linear interpolation to estimate treatment effects at timepoints not reported in comparator studies [35]. In their publication describing a process for assessing NMA feasibility, Cope et al. suggested considering ITC approaches that combine multiple timepoints for outcomes measured over time [37]. Considering only a single timepoint may be insufficient to capture variations in response resulting from varying dosing schedules. Analyzing data at a single timepoint can introduce bias, favoring one treatment over another based on the chosen timepoint. As with ITC methods more broadly, it may be appropriate to conduct ITC analyses for gMG treatments using different timepoint data for CFB outcomes: using a single reported timepoint (e.g., per protocol definition), taking the average across multiple timepoints, or calculating the AUC. For example, an EMA report on zilucoplan included submitted ITCs using AUC for MG-ADL and QMG CFB outcomes, which were described as measuring an aggregated effect over time to facilitate comparison between efgartigimod (with cyclical dosing and effect in ADAPT) and zilucoplan (with continuous dosing and a sustained effect in RAISE) [24]. There is a need for further investigation of the strengths and weaknesses of these approaches for ITCs of gMG treatments, but ultimately it is important to explore multiple approaches to provide a comprehensive view of how differences in timepoints can impact ITC results.

Supporting our finding of substantial cross-trial variation in MG-ADL and QMG placebo response, a relatively high placebo response was also recently reported for the LUMINESCE phase III satralizumab trial and the MINT phase III inebilizumab trial, which were published after the search dates for our SLR [56,57]. Placebo arm heterogeneity and its impact on ITC results has been the focus of previous research with potential implications for ITCs of gMG treatments. Specifically, recent ITC publications have identified unanchored ITCs as appropriate approaches where there is evidence of placebo arm heterogeneity [35,58,59]. Prior to conducting ITCs for treatments for postpartum depression. Meltzer-Brody et al. evaluated the placebo arms of included studies [58]. The authors noted the placebo arm of one trial had a substantially larger CFB for outcomes of interest, which they attributed to more frequent in-person study visits or expectations regarding the therapy [58]. Because of this difference, they determined the placebo arm was not an appropriate common comparator in ITCs and opted to conduct unanchored MAICs [58]. To explore how differences in route of administration affect the results of ITCs where placebo is the common comparator, Fawsitt et al. compared anchored and unanchored ITC methods using CFB outcome data from placebo-controlled RCTs of subcutaneous and intravenous migraine preventative treatments [35]. The authors found results varied based on the ITC method used and concluded that in situations where there is cross-trial heterogeneity in placebo characteristics, decision makers should consider conducting unanchored ITCs in addition to NMA [35]. Fawsitt et al. further noted differences in routes of administration may influence patient perceptions regarding treatment efficacy and thereby affect the size of placebo response for patient-reported outcomes [35]. As such, anchored ITCs based on evidence networks of treatments with different routes of administration may be susceptible to expectation bias [35].

Several ITCs of gMG treatments have recently been published [60–63]. An examination of these publications (e.g., reported limitations) can help inform future ITCs in this therapeutic area. Ma et al. and Chen et al. each conducted NMAs using phase II and III RCTs for both approved and unapproved gMG treatments [60,61]. The authors of both studies acknowledged the small sample sizes of some included trials and cross-trial patient population differences as limitations of their analyses. A commentary on Chen et al. raised concerns regarding their target population not being clearly defined with respect to whether it included AChR+ and/or MuSK+ patients and criticized the lack of discussion around cross-trial differences in TEMs and outcome assessment timepoints [27]. Aligned with our findings, Chen et al. identified rituximab trials as being associated with higher heterogeneity and removed RINOMAX in a sensitivity analysis [60]. Saccà et al. (2023) also recently conducted an exploratory NMA as part of a broader SLR and meta-analysis [62]. The authors noted differences in patient characteristics between trials for rituximab and other treatments but did not investigate the sources of heterogeneity. Other than rituximab, the authors included only registration phase III RCTs without explanation for how trials or treatments were selected. Two recent anchored MAIC analyses compared efgartigimod with ravulizumab and rituximab on MG-ADL CFB. However, neither reported a formal approach to identifying important TEMs and both considered only a single timepoint for analysis [64,65].

Strengths of this study included the use of a systematic review to build an evidence base for the ITC feasibility assessment, the comprehensive assessment of trial characteristics in alignment with best practices, and thorough consideration of the advantages and disadvantages of multiple ITC methods in relation to the challenges posed by the gMG evidence base.

There were also certain limitations with the current study. First, we focused on MG-ADL and QMG outcomes, although other efficacy and safety outcomes are reported for RCTs in gMG. MG-ADL and QMG were considered to be the most relevant because they were the most commonly reported efficacy outcomes in the included RCTs, and further, were often the primary or key secondary outcomes in these trials. They are also widely accepted as the most relevant parameters for MG severity. Second, we identified potential within-trial imbalances and potential TEMs using semi-quantitative approaches that were not validated. Although we evaluated reported RCT subgroup analyses to identify potential TEMs, these varied across trials and so there is uncertainty as to what characteristics are the most important TEMs. Third, some characteristics were not consistently reported across trials, limiting our ability to characterize clinically important heterogeneity and potential impacts on the choice of ITC methods.