Indirect comparisons of siponimod with fingolimod and ofatumumab in multiple sclerosis: assessing the feasibility of propensity score matching analyses

Abstract

Objective

Head-to-head trials comparing siponimod with fingolimod or ofatumumab in patients with multiple sclerosis (MS) are lacking. Instead, the comparative efficacy of siponimod can be derived from indirect treatment comparisons (ITCs). We assessed the suitability of ITCs leveraging individual patient data from relevant phase III trials across different MS phenotypes.

Methods

One siponimod trial in patients with secondary progressive MS (SPMS), four fingolimod trials (three in relapsing-remitting MS [RRMS], and one in primary progressive MS [PPMS]), and two ofatumumab trials in relapsing MS (RMS) were considered. The suitability of ITCs was evaluated based on trial design, patient eligibility criteria, baseline patient characteristics, placebo response, and outcome definitions for each trial. Analyses deemed feasible were conducted using one-to-one propensity score matching (PSM).

Results

An ITC between siponimod in SPMS and either fingolimod in RRMS or ofatumumab in RMS was not feasible because of insufficient overlap in key patient characteristics (e.g. disability level and relapse history) and differences in placebo response. However, a comparison between siponimod in SPMS and fingolimod in PPMS was feasible because of sufficient overlap in eligibility criteria and baseline characteristics. One-to-one PSM demonstrated siponimod was favored relative to fingolimod for time to 6- and 3-month confirmed disability progression though not significantly different (hazard ratio 0.76 [95% confidence interval 0.48–1.20; p-value = .240] and hazard ratio 0.80 [95% confidence interval 0.52–1.22; p-value = .300], respectively).

Conclusions

For trials in MS, clinical phenotype is an important determinant of ITC feasibility. An ITC between siponimod in SPMS and either fingolimod in RRMS or ofatumumab in RMS was not feasible. The only feasible comparison was between siponimod in SPMS and fingolimod in PPMS.

Keywords:

• Comparative efficacy
• disease phenotype
• feasibility assessment
• indirect treatment comparison
• multiple sclerosis
• patient-level data
• primary progressive
• propensity score matching
• relapsing-remitting
• secondary progressive

Introduction

Multiple sclerosis (MS) is the leading cause of disability in young and middle-aged people and affects ∼2.8 million individuals worldwide¹. Several clinical phenotypes have been identified for MS². Relapsing-remitting MS (RRMS) is the most common MS phenotype, affecting 70% to 80% of MS patients, and is characterized by the occurrence, recurrence, or worsening of symptoms of neurologic dysfunction (relapses) followed by complete or partial remission³. Secondary progressive MS (SPMS) is characterized by a gradual neurological deterioration independent of relapses after an initial relapsing-remitting course with or without superimposed relapses³. The term relapsing MS (RMS) describes both RRMS and SPMS with superimposed relapses (i.e. relapsing forms of MS). Primary progressive MS (PPMS) affects 15–20% of MS patients and presents as a gradual deterioration from disease onset with few or no relapses³.

Disease-modifying therapies (DMTs) are the standard of care for people with MS. They reduce both relapse frequency and the underlying acute focal inflammation in the CNS visible on magnetic resonance imaging (MRI) scans, albeit with varying degrees of efficacy⁴. They also reduce the risk of disability progression with limited efficacy⁴. Beta interferons (IFN beta) and glatiramer acetate (GA) collectively comprise the BRACE therapies (Betaseronⁱ, Rebifⁱⁱ, Avonexⁱⁱⁱ, Copaxone^iv, Extavia^v), which were the first DMTs approved, and are still widely used. Although they have relatively benign safety profiles, the injectable BRACE therapies are less effective compared to more modern oral or injectable DMTs^5,6.

Siponimod is an oral sphingosine 1-phosphate receptor modulator studied in SPMS and recently approved with diverse labels, e.g. in the United States (US) (patients with RMS)⁷, the European Union (EU), and Canada (adult patients with SPMS with active disease)^8,9, and Australia (adult patients with SPMS)¹⁰. Approval of siponimod was based on a phase III, double-blind, randomized EXPAND trial that demonstrated oral siponimod significantly reduced the risk of disease progression in a broad SPMS population that included patients with a high level of disability¹¹.

Fingolimod was the first oral sphingosine 1-phosphate modulator approved for MS. Fingolimod is approved for adult and pediatric patients with RMS in the US, Canada (RRMS in adults), and Australia^12–14 and highly active RRMS in the EU¹⁵. Fingolimod has been evaluated in patients with PPMS¹⁶ and RRMS^5,17,18 but not in SPMS patients¹⁸. In the phase III trial in patients with PPMS, disability progression and brain volume change were not different between the fingolimod and placebo groups, although fingolimod was found to improve MRI outcomes relative to placebo¹⁶.

Ofatumumab is a subcutaneous anti-CD20 monoclonal antibody treatment recently approved for adult patients with RMS in the US and Australia^19,20, RMS with active disease in the EU²¹, and RRMS with active disease in Canada²². In two phase III trials, ofatumumab was more effective in reducing focal inflammation (relapses and CNS inflammation as measured by MRI) and delayed disability worsening compared to teriflunomide²⁴.

There are no head-to-head randomized clinical trials (RCTs) directly comparing siponimod to either fingolimod or ofatumumab, so their relative efficacy is unknown. In the absence of direct evidence, indirect treatment comparisons (ITCs) can be used to explore their relative effect if it is feasible to perform valid ITCs based on the available clinical trial evidence, which needs to be assessed. Given their similar listing status in the US, there may be a tendency for researchers to want to include the three therapies in an ITC, such as a network meta-analysis (NMA), which simultaneously compares the relative efficacy of therapies in the absence of direct evidence. However, a recent review of NMAs in MS discussed the importance of the population being similar in terms of MS phenotype when undertaking such analyses²³. Indirectly comparing trials in different MS phenotypes may produce invalid results if there are substantial phenotype-driven differences in patient populations (e.g. age and stage of disease). The pivotal trial for siponimod was conducted in patients with a different MS phenotype compared to trials for fingolimod and ofatumumab, so systematically evaluating cross-trial differences to assess the validity of undertaking ITCs is particularly important to avoid misleading comparisons of treatment effects. We undertook this present work to establish whether there is merit in evaluating siponimod in an MS phenotype for which it had not been directly evaluated, thereby aiming to provide more clarity on caveats regarding the comparability of different MS phenotypes and the clinical implications for healthcare providers and patients.

Propensity score matching (PSM) is a common approach for comparing the efficacy of two or more treatments using observational data^24–27. More recently, PSM has also been used to derive ITCs using individual patient data (IPD) from multiple RCTs evaluating therapies with similar therapeutic indications^25,26. Propensity score methods leverage IPD for both patient populations being compared to correct for differences between populations that are expected to bias comparisons. We considered PSM and other propensity score-based ITC approaches using IPD from pivotal trials evaluating fingolimod, ofatumumab, and siponimod. We were in a unique position with this study because IPD was available for trials of each therapy. Consequently, we could adjust for cross-trial differences to allow for more valid indirect estimates of efficacy.

This study aimed to evaluate the suitability of performing PSM-based ITC analyses involving siponimod and trials in different MS phenotypes by systematically evaluating cross-trial heterogeneity between the EXPAND trial and the phase III trials evaluating fingolimod and ofatumumab. Where feasible, we conducted PSM analyses to evaluate the comparative efficacy of siponimod vs. fingolimod or ofatumumab for the treatment of MS using IPD from the respective pivotal trials for these therapies.

Methods

Data sources

The three pre-specified therapies of interest were siponimod, fingolimod, and ofatumumab. The double-blind, phase III, randomized EXPAND trial compared siponimod to placebo in patients with SPMS¹¹. Four double-blind, phase III, randomized trials compared fingolimod to placebo (FREEDOMS¹⁸ and FREEDOMS II¹⁷ in RRMS) or intramuscular IFN beta-1a (TRANSFORMS⁵ in RRMS, while INFORMS evaluated patients with PPMS¹⁶). The double-blind, phase III, randomized ASCLEPIOS I/II trials were identically designed trials that compared ofatumumab to teriflunomide in patients with RMS²⁸. The design characteristics of the trials are summarized in Table 1.

Table 1. Design of trials included in this study.

Trial name	ClinicalTrials.gov identifier	Pivotal publication (author, year)	Phase	Randomization and blinding	MS population	Sample size	Treatment(s)	Comparator	Trial duration
EXPAND	NCT01665144	Kappos, 2018 [11]	III	Randomized Double-blind	SPMS	1651	Siponimod 2 mg PO QD	Placebo PO QD	Up to 3 years
INFORMS	NCT00731692	Lublin, 2016 [16]	III	Randomized Double-blind	PPMS	970	Fingolimod 1.25 mg PO QD, 0.5 mg PO QD^a	Placebo PO QD	Up to 5 years
FREEDOMS	NCT00289978	Kappos, 2010 [18]	III	Randomized Double-blind	RRMS	1272	Fingolimod 1.25 mg PO QD, 0.5 mg PO QD	Placebo PO QD	2 years
FREEDOMS II	NCT00355134	Calabresi, 2014 [17]	III	Randomized Double-blind	RRMS	1083	Fingolimod 1.25 mg PO QD, 0.5 mg PO QD	Placebo PO QD	2 years
TRANSFORMS	NCT00340834	Cohen, 2010 [5]	III	Randomized Double-blind	RRMS	1280	Fingolimod 1.25 mg PO QD, 0.5 mg PO QD	IFNB-1a IM 30 µg QW	1 year
ASCLEPIOS I	NCT02792218	Hauser, 2020 [28]	III	Randomized Double-blind	RMS (94.1% RRMS and 5.9% active SPMS)	927	Ofatumumab 20 mg SC Q4W^b	Teriflunomide 14 mg PO QD	Up to 30 months
ASCLEPIOS II	NCT02792231	Hauser, 2020 [28]	III	Randomized Double-blind	RMS (94.5% RRMS and 5.5% active SPMS)	955	Ofatumumab 20 mg SC Q4W^b	Teriflunomide 14 mg PO QD	Up to 30 months

Abbreviations. IFNB: interferon beta; MS: multiple sclerosis; PO: oral; PPMS: primary progressive multiple sclerosis; Q4W: every four weeks; QD: daily; QW: weekly; RMS: relapsing multiple sclerosis; RRMS: relapsing-remitting multiple sclerosis; SC: subcutaneous; SPMS: secondary progressive multiple sclerosis.

^a All patients in the fingolimod group were administered 1.25 mg fingolimod at study commencement but switched to 0.5 mg mid-study because of a protocol amendment. All patients assigned to the fingolimod group completed the study on 0.5 mg fingolimod.

^b All patients in the ofatumumab group were initially administered 20 mg loading doses on days 1, 7, and 14.

Ethics

Novartis Pharma AG was the sponsor of all trials of all three pre-specified therapies of interest used in this study. Concept and design for PSM analyses underwent Novartis Pharma AG approval and EVERSANA^vi was provided access to anonymized data from all trials of interest as described previously, to run the analyses after approval. All trials were registered with the U.S. National Library of Medicine ClinicalTrials.gov. Trial protocols were institutional review board- or independent ethics committee-approved. Trials were conducted by the International Conference on Harmonization Guidelines for Good Clinical Practice and the principles of the Declaration of Helsinki.

Feasibility assessment

Systematic differences between trials may compromise the validity of ITCs, therefore the degree of heterogeneity in patient characteristics and trial design between trials was assessed. A qualitative assessment of between-trial heterogeneity based on the trial design, patient eligibility criteria, baseline patient characteristics, placebo arm outcome values (i.e. placebo response), and outcome definitions was performed to ensure similarity. The exchangeability assumption underlying each potential ITC was assessed by comparing outcomes for placebo arms between clinical trials, where applicable.

The following study characteristics were extracted for each trial if reported: general publication details (e.g. author, study location, year, journal, and NCT number), trial design (e.g. phase, blinding, allocation, and trial duration), patient eligibility criteria, baseline patient characteristics (e.g. age, sex, Expanded Disability Status Scale [EDSS] score, and duration of disease), intervention details (e.g. therapy, dose, route, and frequency), and trial-specific outcome definitions for the available outcomes of interest. We used IPD for each trial to evaluate standardized mean differences (SMDs) and the overlap of key baseline patient characteristics between trials.

Where feasible, ITCs were conducted using PSM to evaluate the comparative efficacy of siponimod vs. fingolimod or ofatumumab for the treatment of MS using IPD from pivotal trials for these therapies. Methodological details for the PSM analyses are provided in Supplemental Appendix A.

Outcomes of interest

Outcomes of interest (selected a priori) were time to 6-month confirmed disability progression (6mCDP), time to 3-month CDP (3mCDP), and annualized relapse rate (ARR), which were primary or secondary endpoints evaluated in the EXPAND, FREEDOMS, FREEDOMS II, INFORMS (CDP outcomes only), TRANSFORMS, and ASCLEPIOS I/II trials. These endpoints are clinically important and commonly assessed in MS trials. Notably, ARR was not reported in INFORMS, which was conducted in a PPMS population with no or very few relapses, and time to 6mCDP was not reported in TRANSFORMS, which was a one year study and so not suited to measuring this outcome.

Results

Feasibility assessment

Trial design

All included trials were double-blind, parallel-group, phase III RCTs (Table 1). Trials had either a placebo, IFN beta, or teriflunomide as comparators, as described previously. Trial durations ranged from one year (TRANSFORMS) up to five years (INFORMS). Trial populations included RRMS (fingolimod), RMS (ofatumumab), SPMS (siponimod), and PPMS (fingolimod). In the two ofatumumab trials, most patients had RRMS (∼94%), and the remainder presented with active SPMS (∼6%).

Patient eligibility criteria

The results of a qualitative comparison of patient eligibility criteria for the included trials are summarized in Table 2. All trials included adult patients; however, the trials differed in inclusion/exclusion criteria regarding MS phenotype evaluated, baseline EDSS score, history of relapse or RRMS, and history of disease progression. All trials in relapsing MS enrolled patients with an EDSS score between 0.0 and 5.5, but the two trials in progressive MS required patients to have an EDSS score between 3.0 and 6.5 (EXPAND) or between 3.5 and 6.0 (INFORMS). Notably, the majority (56%) of the patients in EXPAND (in SPMS) had a baseline EDSS of ≥6.0, and so would have been excluded from the RRMS/RMS trials. Although EXPAND and all the included RRMS/RMS trials required history or a current diagnosis of RRMS, INFORMS excluded patients with a previous history of relapse and RRMS. Finally, EXPAND and INFORMS required documented progression in the prior two years. The ASCLEPIOS I/II trials also included patients with SPMS with disease activity in addition to patients with RRMS.

Table 2. Comparison of patient eligibility criteria for included trials.

Trial name	Inclusion criteria
	MS type	MS diagnostic criteria	Age	Baseline EDSS score	Required documented progression in prior years	Required absence of relapses in months prior	History of RRMS	Prior DMT and corticosteroid use	Required T25FW test score	Disease duration
EXPAND	SPMS	2010 revised McDonald criteria	18–60	3.0–6.5	Documented EDSS progression in the 2 years before the study ≥1.0 point for patients with EDSS < 6.0 ≥0.5 point for patients with EDSS ≥ 6.0 at screening	No relapse in 3 months prior	History of RRMS	Prior IFN or GA: allowed Prior corticosteroid: disallowed within 3 months Multiple additional criteria regarding specific prior DMTs	NR	Documented SPMS ≥6 months before study enrollment (and preceding history of RRMS)
INFORMS	PPMS	2005 revised McDonald criteria	25–65	3.5–6.0	≥1 year of disease progression^a Documented EDSS progression in the 2 years before the study ≥0.5 point for all patients	No history of relapses allowed	No history of RRMS allowed	Prior IFN or GA: disallowed within 3 months Prior corticosteroid: disallowed within 3 months Multiple additional criteria regarding specific prior DMTs	<30 s	Time since reported PPMS symptoms between 2 and 10 years
FREEDOMS	RRMS	2005 revised McDonald criteria	18–55	0.0–5.5	n/a – not progressive MS	No relapse in 1 month prior	Current diagnosis of RRMS	Prior IFN or GA: disallowed within 3 months Prior corticosteroid: Disallowed within 30 days	NR	NR
FREEDOMS II	RRMS	2005 revised McDonald criteria	18–55	0.0–5.5	n/a – not progressive MS	No relapse in 1 month prior	Current diagnosis of RRMS	Prior IFN or GA: disallowed within 3 months Prior corticosteroid: disallowed within 30 days	NR	NR
TRANSFORMS	RRMS	2005 revised McDonald criteria	18–55	0.0–5.5	n/a – not progressive MS	No relapse in 1 month prior	Current diagnosis of RRMS	Prior IFN or GA: Allowed Prior corticosteroid: disallowed within 30 days	NR	NR
ASCLPEIOS I	RMS	2010 revised McDonald criteria	18–55	0.0–5.5	Not required for SPMS patients	Neurologically stable within 1 month prior	Current diagnosis of RMS	Prior IFN or GA: allowed Prior corticosteroid: disallowed within 30 days Multiple additional criteria regarding specific prior DMTs	NR	≤10 years in patients with EDSS score ≤2.0
ASCLEPIOS II	RMS	2010 revised McDonald criteria	18–55	0.0–5.5	Not required for SPMS patients	Neurologically stable within 1 month prior	Current diagnosis of RMS	Prior IFN or GA: allowed Prior corticosteroid: disallowed within 30 days Multiple additional criteria regarding specific prior DMTs	NR	≤10 years in patients with EDSS score ≤2.0

Abbreviations. CSF: cerebrospinal fluid; DMT: disease-modifying therapy; EDSS: Expanded Disability Status Scale; GA: glatiramer acetate; Gd+: gadolinium-enhancing; IFN: interferon; MRI: magnetic resonance imaging; MS: multiple sclerosis; n/a: not applicable; NR: not reported; PPMS: primary progressive multiple sclerosis; RMS: relapsing multiple sclerosis; RRMS: relapsing-remitting multiple sclerosis; SPMS: secondary progressive multiple sclerosis; T25FW: timed 25-foot walk.

^aPlus two of the following: positive brain MRI, positive spinal cord MRI, positive CSF.

Baseline patient characteristics

Baseline patient characteristics for the included trials are summarized in Table 3. Differences in disease-related baseline characteristics (e.g. EDSS score, duration of MS symptoms and disease, and relapse history) reflected the MS phenotypes studied. Baseline characteristics were substantially different between the siponimod arm in SPMS and the fingolimod and ofatumumab arms in RRMS or RMS trials, respectively. Differences were present but less pronounced in SPMS and PPMS trials. Density plots from the siponimod, fingolimod, or ofatumumab arm were generated using IPD to visualize the degree of overlap between the trial patient populations (Figure 1 and Supplemental Appendix B). Siponimod-treated SPMS patients more closely resembled PPMS patients treated with fingolimod than RRMS/RMS patients treated with fingolimod or ofatumumab. As expected, baseline EDSS score and relapse history characteristics demonstrated a high degree of heterogeneity between patients with SPMS and patients with either RRMS or RMS for the included trials. The two populations (SPMS and RRMS/RMS) also differed substantially in terms of age, duration of MS, and proportion with T1 gadolinium-enhancing (Gd+) lesions at baseline. In contrast to the differences noted between SPMS and RRMS/RMS populations, baseline characteristics including EDSS score, relapse history, the proportion with Gd + lesions, and age were more similar for SPMS and PPMS patients vs. patients with RRMS/RMS.

Figure 1. Distribution of baseline patient characteristics across trials in different MS phenotypes: EXPAND (SPMS), FREEDOMS (RRMS), and INFORMS (PPMS). Note: The INFORMS trial was not included in the number of relapses plots because the PPMS patients did not have relapses. For the time since first MS symptom and time since MS diagnosis, comparisons with INFORMS (PPMS) may be biased by the systematic underestimation of MS duration in PPMS patients (who typically have accumulated lesions but not acute neurological events). Additional comparisons between trials included in this study can be found in Supplemental Appendix B. Abbreviations. EDSS: Expanded Disability Status Scale; MS: multiple sclerosis; PPMS: primary progressive multiple sclerosis; RRMS: relapsing-remitting multiple sclerosis; SPMS: secondary progressive multiple sclerosis.

Table 3. Published patient baseline characteristics for the siponimod, fingolimod (0.5 mg), or ofatumumab treatment arm of included trials.

Characteristic	EXPAND (SPMS) (n = 1105)	INFORMS (PPMS) (n = 336)	FREEDOMS (RRMS) (n = 425)	FREEDOMS II (RRMS) (n = 358)	TRANSFORMS (RRMS) (n = 431)	ASCLEPIOS I (RMS) (n = 465)	ASCLEPIOS II (RMS) (n = 481)
Age (mean years)	48.0	48.5	36.6	40.6	36.7	38.9	38.0
Proportion female (%)	61%	49%	70%	77%	65%	68%	66%
EDSS score (mean)	5.4	4.7	2.3	2.4	2.2	3.0	2.9
EDSS score (median)	6.0	4.5	2.0	2.0	2.0	3.0	3.0
Time since first MS symptom (mean years)	17.1	5.8	8.0	10.4	7.5	8.4	8.2
Time since MS diagnosis (mean years)	12.9	2.8	4.8	6.0	4.8	5.8	5.6
Time since conversion to SPMS (mean years)	3.9	N/A	N/A	N/A	N/A	N/A	N/A
Proportion of patients with Gd + lesions on T1 weighted images (%)	21%	14%	38%	39%	33%	37%	44%
Total volume of T2 lesions (mean mm^3)	15 632	9443	6128	5484	5170	13 168	14 254
Normalized brain volume (cm^3)	1422	1491	1521	1522	1524	1439	1440
Time since most recent relapse (mean years)	5.2	N/A	0.52	0.58	0.44	0.59	0.65
Proportion of patients who were relapse-free in prior year (%)	79%	100%	2%	3%	3%	6%	6%
Proportion of patients who were relapse-free in prior 2 years (%)	64%	100%	0%	0%	0%	2%	3%
Number of relapses per patient in the previous year (mean)	0.2	0	1.5^a	1.4	1.5	1.2	1.3
Number of relapses per patient in the previous 2 years (mean)	0.7	0	2.1	2.2	2.3	2.1	2.0
Proportion of patients without previous use of a DMT (%)	22%	81%	60%	26%	45%	41%	41%

Abbreviations. DMT: disease-modifying therapy; EDSS: Expanded Disability Status Scale; Gd+: gadolinium-enhancing; MS: multiple sclerosis; N/A: not applicable; PPMS: primary progressive multiple sclerosis; RMS: relapsing multiple sclerosis; RRMS: relapsing-remitting multiple sclerosis; SPMS: secondary progressive multiple sclerosis.

Outcome definitions

The outcome definitions for time to 6mCDP, time to 3mCDP, and ARR for each included trial are summarized in Table 4. For each of the reported outcomes, the definitions were described similarly across the trials.

Table 4. Summary of trial-specific outcome definitions.

EXPAND	INFORMS	FREEDOMS	FREEDOMS II	TRANSFORMS	ASCLEPIOS I/II
Time to 6mCDP: 1.0-point increase in EDSS score if the baseline score was 3.0–5.0 0.5-point increase in EDSS score if the baseline score was 5.5–6.5	Time to 6mCDP: 1.0-point increase in EDSS score if the baseline score was ≤5.0 0.5-point increase in EDSS score if the baseline score was ≥5.5	Time to 6mCDP: 1.0-point increase in EDSS scores if the baseline score was <5.5 0.5-point increase in EDSS scores if the baseline score was 5.5	Time to 6mCDP: 1.0-point increase in EDSS score if the baseline score was ≤5.0 0.5-point increase in EDSS score if the baseline score was >5.0	Not available	Time to 6mCDP: 1.5-point increase in EDSS score if the baseline score was 0.0 1.0-point increase in EDSS score if the baseline score was 0.5–5.0 0.5-point increase in EDSS score if the baseline score was >5.0
Time to 3mCDP: 1.0-point increase in EDSS score if the baseline score was 3.0–5.0 0.5-point increase in EDSS score if the baseline score was 5.5–6.5	Time to 3mCDP: 1.0-point increase in EDSS score if the baseline score was ≤5.0 0.5-point increase in EDSS score if the baseline score was ≥5.5	Time to 3mCDP: 1.0-point increase in EDSS score if the baseline score was <5.5 0.5-point increase in EDSS score if the baseline score was 5.5	Time to 3mCDP: 1.0-point increase in EDSS score if the baseline score was ≤5.0 0.5-point increase in EDSS score if the baseline score was >5.0	Time to 3mCDP: 1.0-point increase in EDSS score if the baseline score was ≤5.0 0.5-point increase in EDSS score if the baseline score was ≥5.5	Time to 3mCDP: 1.5-point increase in EDSS score if the baseline score was 0.0 1.0-point increase in EDSS score if the baseline score was 0.5–5.0 0.5-point increase in EDSS score if the baseline score was >5.0
ARR: Average number of confirmed relapses per year (the total number of confirmed relapses divided by the total number of days in the study multiplied by 365.25)	ARR: Not applicable	ARR: Average number of confirmed relapses per year (the total number of confirmed relapses divided by the total number of days on study multiplied by 365.25)	ARR: Average number of confirmed relapses per year (the total number of confirmed relapses divided by the total number of days on study multiplied by 365.25)	ARR: The number of confirmed relapses during a 12-month period	ARR: The number of confirmed relapses of multiple sclerosis per year
Relapse: Appearance of a new neurological abnormality^a Separated by at least 30 days from the onset of a preceding clinical demyelinating event Abnormality must have been present for at least 24 h Occurs in the absence of fever or infection	Relapse: Not applicable	Relapse: Appearance of a new or worsening neurological abnormality^a Separated by at least 30 days from onset of a preceding relapse Abnormality must be present for at least 24 h Occurs in the absence of fever or infection	Relapse: Appearance of a new neurological abnormality^a Separated by at least 30 days from the onset of a preceding clinical demyelinating event Abnormality must have been present for at least 24 h Occurs in the absence of fever or infection Accompanied by an increase of at least 0.5 points on the EDSS score, or an increase of at least 1 point on two FS scores, or an increase of at least 2 points on one FS score^b	Relapse: New, worsening, or recurrent neurological symptoms Occurs at least 30 days after the onset of preceding relapse Duration of at least 24 h without fever or infection Accompanied by an increase of ≥0.5-point on the EDSS or an increase of ≥ 1-point in two FS scores or an increase of ≥ 2-points in one FS score^b	Relapse: Appearance of a new or worsening neurological abnormality^a Separated by at least 30 days from onset of a preceding clinical demyelinating event Abnormality must have been present for at least 24 h Occurs in the absence of fever or known infection Accompanied by an increase of at least 0.5 points on the EDSS score, or an increase of at least 1 point on at least two FS scores, or an increase of at least 2 points on at least one FS score^b

Abbreviations. 3mCDP: 3-month confirmed disability progression; 6mCDP: 6-month confirmed disability progression; ARR: annualized relapse rate; EDSS: Expanded Disability Status Scale; FS: functional system; IPD: individual patient data.

Note. INFORMS and TRANSFORMS did not report time to 6mCDP in their pivotal publications. However, time to 6mCDP outcome data for INFORMS was obtained from the IPD and so the definition of 6mCDP was assumed to reflect the published definition of 3mCDP. This outcome was not available from the TRANSFORMS IPD.

Note. INFORMS did not report ARR.

Note. In all trials, disability progression had to be confirmed at a scheduled visit at least 6 months later (time to 6mCDP) or at least 3 months later (time to 3mCDP).

^aIncludes worsening of the previous stable or improving pre-existing neurological abnormalities.

^bExcluding changes in bowel or bladder function and cognition.

Placebo response

All trials except TRANSFORMS and ASCLEPIOS I/II were placebo-controlled (Table 1), which allowed for assessment of placebo response for time to 6mCDP and 3mCDP endpoints. Placebo response was significantly different between EXPAND and the placebo-controlled trials in RRMS/RMS. Although more similar to one another than to the trials in RRMS/RMS, there were observable differences in placebo response (i.e. placebo arm time to CDP rates) between EXPAND and INFORMS, highlighting differences in the baseline characteristics of SPMS (EXPAND) and PPMS (INFORMS) populations.

There was considerable overlap between the placebo arm values of EXPAND and INFORMS for time to 6mCDP and 3mCDP (Figure 2). However, there was a significantly higher rate of 6mCDP events and a trend toward a higher rate of 3mCDP in INFORMS compared to EXPAND. The HR (95% confidence interval [CI]; p-value) for EXPAND vs. INFORMS was 0.79 (0.63–0.99; p = .0383) and 0.85 (0.69–1.04; p = .103) for time to 6mCDP and 3mCDP, respectively.

Figure 2. Time to confirmed disability progression in placebo arms of the EXPAND, FREEDOMS, FREEDOMS II, and INFORMS trials. Note: The shaded areas of the plot that surround the curves represent the 95% confidence intervals around the point estimates at each time point. Curves were derived from IPD for the placebo arms of the indicated trials: EXPAND (SPMS), FREEDOMS (RRMS), FREEDOMS II (RRMS), and INFORMS (PPMS). Curves were generated before excluding relapsing patients from EXPAND and before conducting PSM on treatment arms (EXPAND and INFORMS). The TRANSFORMS (in RRMS) and ASCLEPIOS I/II (in RMS) trials were not included because they were not placebo-controlled. Abbreviations. 3mCDP: 3-month confirmed disability progression; 6mCDP: 6-month confirmed disability progression; IPD: individual patient data; PPMS: primary progressive multiple sclerosis; PSM: propensity score matching; RMS: relapsing multiple sclerosis; RRMS: relapsing-remitting multiple sclerosis; SPMS: secondary progressive multiple sclerosis.

As expected, considerable overlap in placebo arm values was also noted between FREEDOMS and FREEDOMS II for time to 6mCDP and 3mCDP. However, between EXPAND and FREEDOMS/FREEDOMS II, a distinct difference was evident as shown by the separation of Kaplan-Meier curves for these pairs of trials (Figure 2). The fingolimod trials in RRMS, where CDP often results from incomplete relapse recovery, each demonstrated a significantly lower rate of CDP events compared to the EXPAND trial in SPMS. The risk of 6mCDP events was significantly different in the placebo arms of EXPAND vs. FREEDOMS (HR 1.59 [95% CI 1.20–2.11; p = .00133]) and EXPAND vs. FREEDOMS II (HR 1.73 [95% CI 1.26–2.35; p < 0.001]). Similarly, the risk of 3mCDP events was significantly different in the placebo arms of EXPAND vs. FREEDOMS (HR 1.57 [95% CI 1.22–2.02; p < 0.001]) and EXPAND vs. FREEDOMS II (HR 1.32 [95% CI 1.02–1.70; p = .0357]).

Comparison of placebo response was not possible for TRANSFORMS and ASCLEPIOS I/II because of the lack of placebo arms for these trials.

Propensity score matching

An ITC between siponimod (EXPAND, in SPMS patients) and fingolimod (INFORMS, in PPMS patients) was feasible. Sufficient overlap in trial design, patient eligibility criteria, baseline patient characteristics, outcome definitions, and placebo response, as described previously, permitted the use of IPD to estimate the relative efficacy of the interventions within matched cohorts.

Patient selection and population alignment

Of the 1044 siponimod-treated SPMS patients in EXPAND, 366 patients with relapses in the two years before study entry were removed before the calculation of propensity scores to align more closely with the eligibility criteria in INFORMS. In total, 678 patients treated with siponimod 2 mg in EXPAND and 332 patients treated with fingolimod 0.5 mg in INFORMS were considered for the PSM analysis. Not all possible covariates could be matched (Supplemental Appendix C). After matching on seven covariates (age, baseline EDSS score, time since first MS symptom, timed 25-foot walk test, normalized brain volume, sex, and volume of T2 lesions), a total of 117 patients from each treatment group were included in the time to 6mCDP and time to 3mCDP analyses.

Before matching (and after excluding patients with prior relapses from EXPAND), six of the seven ranked covariates were imbalanced between treatment groups, with SMDs ≥0.20. In particular, matched SPMS patients who received siponimod had on average a long time since first MS symptom (SMD = 2.039), a lower normalized brain volume (SMD = 0.812), and a higher EDSS score (SMD = 0.673) at baseline than PPMS patients who received fingolimod 0.5 mg. Matching improved the balance across treatment groups, with SMDs <0.20. Descriptive statistics for all available patient characteristics before and after matching are presented in Supplemental Appendix C.

Before matching, the estimated probabilities of receiving siponimod, as predicted from baseline patient characteristics, were systematically higher among patients who received siponimod than those who received fingolimod 0.5 mg. After matching, the distribution of predicted probabilities of receiving siponimod was similar across treatment groups. Density plots of the propensity scores estimates for each treatment group before and after matching are presented in Supplemental Appendix C.

Time to CDP

Results of the one-to-one PSM analysis between siponimod and fingolimod are presented in Table 5. Although there was no statistically significant difference, a trend toward a greater delay in time to 6mCDP and 3mCDP favored siponimod compared to fingolimod (HR 0.76 [95% CI 0.48–1.20; p = .240] and HR 0.80 [95% CI 0.52–1.22; p = .300], respectively).

Table 5. Propensity score matching analysis results.

Pre-PSM population			Post-PSM population			PSM ITC results
Siponimod (N)	Fingolimod 0.5 mg (N)	Mean SMD	Siponimod (N)	Fingolimod 0.5 mg (N)	Mean SMD	Outcome	Siponimod vs. Fingolimod 0.5 mg HR (95% CI)
678	332	0.383	117	117	0.094	Time to 6mCDP	0.76 (0.48–1.20)
						Time to 3mCDP	0.80 (0.52–1.22)

Abbreviations. 3mCDP: 3-month confirmed disability progression; 6mCDP: 6-month confirmed disability progression; CI: confidence interval; HR: hazard ratio; ITC: indirect treatment comparison; N: sample size; PSM: propensity score matching; SMD: standardized mean difference.

Note. HRs are doubly robust. Without being doubly robust, the ITC results were 0.84 (0.54–1.31) and 0.87 (0.58–1.31) for time to 6mCDP and 3mCDP, respectively.

Note. SMD was calculated as the mean absolute difference in a baseline covariate between two comparison groups (i.e. siponimod and fingolimod) divided by the pooled standard deviation of the covariate for the entire analysis sample. The mean SMD represents the average SMD across all ranked characteristics.

Note. The “pre-PSM population” excludes relapsing patients from EXPAND and does not include patients with missing values in covariates required for the PSM analysis.

Scenario analyses wherein the lowest-ranked covariates were dropped from the one-to-one PSM analysis in a stepwise fashion (i.e. reducing the number of characteristics adjusted for) produced similar results as the primary analysis (Supplemental Appendix C). Similarly, additional analyses using other matching and weighting methods (i.e. three-way nearest neighbor matching and multinomial inverse probability of treatment weighting) produced results that were consistent with the primary one-to-one PSM analysis (Supplemental Appendix C).

Discussion

Before conducting ITCs, it is important to assess their feasibility based on the available clinical trial evidence. The validity of ITCs is dependent on whether there are systematic differences between trials, especially for patient characteristics that are treatment effect modifiers^29–32. We identified major differences in patient eligibility criteria, baseline patient characteristics, and placebo response that limited the feasibility of using PSM to conduct valid ITCs between siponimod in SPMS patients and either fingolimod in RRMS or ofatumumab in RMS. There was insufficient overlap in key patient characteristics, such as level of disability and relapse history. However, these characteristics were more similar in SPMS and PPMS patients as compared to RRMS/RMS patients, reflecting the relatively higher level of disability, lower rate of relapse (or absence of relapses), older age, and differing MRI inflammatory disease activity that characterize the progressive MS phenotypes. Relative to patients with progressive MS, RRMS/RMS patients are typically younger and have less advanced disease^33,34. It was therefore feasible to use PSM to conduct valid ITCs between siponimod and fingolimod in patients with progressive MS. The results of our feasibility assessment support the separation of data for trials in different MS phenotypes when conducting ITCs, except in the case of comparing SPMS (EXPAND) and PPMS (INFORMS) using population-based methods to adjust for imbalances. As expected, even though there was some overlap between RRMS/RMS patients and SPMS patients in their baseline level of disability, patients with the same baseline EDSS score but a different MS phenotype (i.e. RRMS/RMS vs. SPMS) were still likely to differ in their relapse history, disease duration, and age.

In addition to patient characteristics, placebo response provides an indication of the degree of comparability in the trial populations. To meet the assumption of exchangeability (i.e. the patients could have been included in either trial in a pairwise ITC), the placebo response of the comparator trial populations must be similar. Although more similar to one another than to the trials in RRMS/RMS, there were observable differences in placebo response between EXPAND and INFORMS that highlight the differences in the SPMS (EXPAND) and PPMS (INFORMS) populations before conducting adjusted analyses.

The importance of conducting a detailed assessment of cross-trial heterogeneity when undertaking ITCs was recently demonstrated by Samjoo et al. for trials in SPMS patients³⁵. In this case, the feasibility assessment highlighted the appreciable cross-trial heterogeneity that would have caused biased results if traditional ITCs, such as Bucher ITCs or NMA were conducted. Instead, population-based ITC methods that accounted for disparities in clinical covariates by leveraging IPD were found to be a more valid approach to examining comparative efficacy. Previous studies in MS and several other therapeutic areas have provided insights into the impact of adjustments on the validity of ITCs^35–38. In these examples, the clinical interpretation of the comparative efficacy of interventions was altered when accounting for differences in clinically relevant covariates.

The results of our ITC feasibility assessment underscore the importance of systematically evaluating cross-trial differences and identifying whether there is a need for covariate adjustment via PSM or comparable methodologies. Given the magnitude of the between-trial differences we observed, it was not methodologically appropriate to include SPMS (EXPAND) in a network of RRMS/RMS trials (FREEDOMS, FREEDOMS II, TRANSFORMS, and ASCLEPIOS I/II). However, it was feasible to indirectly compare trials in SPMS (EXPAND) and PPMS (INFORMS) using population-based methods to adjust for imbalances. These findings support the separation of data for trials in relapsing MS and progressive MS when conducting ITCs. From a clinical standpoint, it is important to exercise caution when interpreting the results of ITCs of trials in different MS phenotypes. People with relapsing MS cannot be readily compared to those with progressive MS. Notably, MS phenotype definitions can differ in clinical practice and between RCTs. For example, different SPMS definitions based on measures of progression and relapse history have been used by clinicians^2,39. Our findings regarding the feasibility of ITCs of trials in different MS phenotypes may not be generalizable to other indirect comparisons if the MS phenotype definitions differ for the included trials.

For MS trials, eligibility criteria related to disease phenotype can substantially influence how similar patient populations are and thus whether ITCs are feasible. Supporting the findings of this study, the importance of MS phenotype as a determinant of ITC feasibility has been noted in other recently published ITCs and feasibility assessments for ITCs. Samjoo et al.³⁵ concluded that ITCs were not feasible between the EXPAND siponimod trial in SPMS patients and the OPERA I/II ocrelizumab trials in RMS patients⁶ in part because the patient populations were too dissimilar. Substantial differences in baseline patient characteristics between EXPAND and OPERA I/II were noted for age, EDSS score, time since onset of MS symptoms, exposure to prior DMTs, prior relapse history, and MRI lesion measures. We observed a similar level of heterogeneity between EXPAND and ASCLEPIOS I/II, which is unsurprising given that the ASCLEPIOS I/II trials, like the OPERA I/II trials, were conducted in RMS patients with a low mean EDSS score (i.e. <3.0) at baseline. Unlike our study, Samjoo et al. did not have access to IPD for the comparator trials, which prevented a specific investigation of patients with progressive disease³⁵. In NMAs by the Institute for Clinical and Economic Review⁴⁰ and Melendez-Torres et al.⁴¹, the authors conducted separate analyses for trials in patients with RRMS and progressive disease phenotypes (PPMS or SPMS). Other recent ITCs in MS have focused on a single MS phenotype^42–46. In agreement with the findings of our study, Filippini et al.⁴⁷ considered RRMS separately in an NMA from progressive disease phenotypes, but combined trial data for patients with SPMS, progressive-relapsing MS, and PPMS, reasoning that these MS phenotypes were sufficiently similar because the course of MS is similar after the onset of progression regardless of prior history.

In the absence of randomized head-to-head comparative clinical trial data, and provided it is appropriate given the evidence base, PSM is a valuable method that enables a pairwise comparison of two treatments from separate trials using IPD. In addition to more broadly assessing the feasibility of ITCs for trials in patients with different MS phenotypes, this study used PSM to generate matched comparisons of siponimod (EXPAND) in SPMS and fingolimod (INFORMS) in PPMS, permitting an assessment of their relative efficacy in terms of time to CDP in patients with these MS phenotypes. After matching, treatment with siponimod was associated with a greater delay in disability progression, though this did not reach statistical significance. Findings were also robust across additional analyses using alternative propensity score methods for both CDP outcomes. Notably, the PSM analysis was a post-hoc analysis not included in statistical analysis plans for RCTs considered in this study and so was not planned to be powered to detect statistical significance.

A notable strength of the present study is that we were able to leverage IPD from multiple RCTs evaluating several DMTs in various MS phenotypes. The availability of IPD allowed us to conduct a rigorous assessment of the comparative efficacy between siponimod and fingolimod and between siponimod and ofatumumab in the absence of a head-to-head trial. Although ITCs involving methods, such as NMA can be performed using aggregate data, the results of these analyses may be biased by trial-level heterogeneity. In contrast, IPD provides more flexibility in adjusting for heterogeneity using standard covariate adjustment, matching, or reweighting techniques.

Our study had several limitations. First, we were unable to examine placebo responses for three of the trials because they did not include placebo arms, and there may be additional trial and patient characteristics that were not captured by the trials but influenced patient outcomes. Second, although we used PSM to adjust for imbalances in demographics and clinical characteristics between trials, ITCs using PSM may be confounded by unobserved but relevant differences between the two groups. To account for this, covariates available in both trials were identified and ranked based on prognostic importance a priori, which enabled adjustment of seven (out of nine) factors between the treatment groups. The two outstanding covariates (duration of SPMS and years since most recent relapse) did not apply to INFORMS because this study was in a PPMS population, which has no or very few relapses. Third, the relatively small number of patients informing the PSM comparison of siponimod with fingolimod reduced the statistical power and effect size for this analysis. Finally, we excluded SPMS patients with relapses (in the two years before study entry) in EXPAND from the PSM analysis. Although removing these patients reduced the number available for matching, it improved the alignment between EXPAND and INFORMS in terms of patient eligibility criteria (i.e. no relapsing MS).

Conclusion

For trials in MS, clinical phenotype is a key determinant of ITC feasibility. In this study, the SPMS phenotype (chronic progression of disability with or without overlaid relapses) as assessed in EXPAND sufficiently resembled the PPMS phenotype (progressive disease from onset) as assessed in INFORMS, permitting the estimation of comparative efficacy using PSM. For time to 6mCDP and 3mCDP, siponimod was favored relative to fingolimod though not significantly different (HR 0.76 [95% CI 0.48–1.20; p = .240] and HR 0.80 [95% CI 0.52–1.22; p = .300], respectively). However, trials evaluating other MS phenotypes (RRMS in FREEDOMS, FREEDOMS II, and TRANSFORMS; and RMS in ASCLEPIOS I/II) were too dissimilar to compare with SPMS, making comparative analyses infeasible. Our findings underscore the importance of assessing population differences across MS trials evaluating different MS phenotypes, as disparities may be too great for even population-based adjustment techniques to overcome and allow for valid comparisons of comparative efficacy. As such, comparative efficacy analyses incorporating different MS phenotypes, without adjustment, should be interpreted with caution.

Transparency

Declaration of funding

This research was supported by Novartis Pharma AG.

Declaration of financial/other relationships

IAS, EW, AH, PS, CD, and CC are paid employees of EVERSANA, which was contracted by Novartis Pharma AG to work on this project. CC has disclosed is also a shareholder of EVERSANA. RB and NA are salaried employees of Novartis Pharma AG, which is the manufacturer of fingolimod, ofatumumab, and siponimod. FD was a salaried employee of Novartis Pharma AG at the time of writing.

Author contributions

IAS, EW, AH, PS, CD, and CC were involved with study concept and design, analysis/interpretation of data, and drafting/revising of the manuscript. IAS, EW, and CC provided study supervision. IAS and CC critically reviewed the importance of intellectual content for the work. RB, FD, and NA assisted with the interpretation of the data and critically reviewed the importance of intellectual content for the work. All authors approved the final version of the manuscript.

Acknowledgements

The authors thank Melody Zhao (employee of EVERSANA) for her contributions to the feasibility assessment and editorial assistance with this manuscript.

Notes

i Betaseron, Bayer, Leverkusen, Germany.

ii Rebif, Merck Serono, Darmstadt, Germany.

iii Avonex, Biogen, Cambridge, MA, USA.

iv Copaxone, Teva Neuroscience, Kansas City, MO, USA.

v Extavia, Novartis, Basel, Switzerland.

vi EVERSANA, Milwaukee, WI, USA.