Abstract
Background
Target trial emulation (TTE) offers a formal framework for causal inference using observational data, but its validity must be evaluated in each research domain by replicating randomised clinical trials (RCTs). We aimed to replicate eight RCTs evaluating the efficacy of disease-modifying therapies (DMTs) in multiple sclerosis (MS) using French registry data.
Methods
This multicentre, retrospective, observational study was conducted using data extracted in December 2023 from the Observatoire Français de la Sclérose en Plaques (OFSEP) database. For each emulated trial, patients were included when they initiated one of the DMT evaluated in the corresponding RCT and met its inclusion criteria. Clinical outcomes were the annualised relapse rate and 3-month confirmed Expanded Disability Status Scale progression. Radiological outcomes were new/enlarged T2-lesions and new gadolinium-enhanced T1-lesions on a brain MRI. A targeted maximum likelihood estimator was used to estimate the treatment effect adjusted for confounding factors between groups and corrected for censoring and missing outcome assessment.
Results
14 111 patients were included in eight emulated trials: ASSESS (fingolimod vs glatiramer acetate), BEYOND (interferon beta vs glatiramer acetate), CONFIRM (dimethyl fumarate (DMF) vs glatiramer acetate), OPERA (ocrelizumab vs interferon beta), REGARD (interferon beta vs glatiramer acetate), RIFUND-MS (rituximab vs DMF), TENERE (teriflunomide vs interferon beta) and TRANSFORMS (fingolimod vs interferon beta). Treatment effects estimated in emulated trials were concordant with RCT findings in seven of eight trials for relapse rate, and in all six trials assessing disability progression. Radiological outcomes were more challenging to replicate; concordance was achieved in three of five trials for new T2-lesions, and one of four trials for new gadolinium-enhanced T1-lesions.
Conclusion
The combined use of a TTE methodology and high-quality registry data is a valid tool to evaluate treatment effectiveness in MS.
Keywords: MULTIPLE SCLEROSIS, NEUROEPIDEMIOLOGY, STATISTICS
WHAT IS ALREADY KNOWN ON THIS TOPIC
Target trial emulation (TTE) provides a methodological framework for causal inference based on observational data, and its validation by replicating randomised clinical trial (RCT) results is essential in every research domain.
WHAT THIS STUDY ADDS
In this study including 14 111 patients from the French multiple sclerosis (MS) registry, we successfully replicated the results of 7 out of 8 RCTs regarding treatment effect on relapse rate in MS, and of all six RCTs assessing disability progression. Radiological outcomes were more challenging to replicate, with concordant results in fewer emulated trials.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
This study supports the validity of TTE using high-quality registry data to provide reliable estimates of treatment effect in MS.
INTRODUCTION
Target trial emulation (TTE) provides a methodological framework essential to answer causal questions using observational data.1 2 It establishes explicit guidelines to ensure the rigorous design of all the key elements of a study protocol: eligibility criteria, treatment strategies and assignment procedures, time zero and follow-up period, outcome definition, causal contrast and statistical analysis plan. Thus, the TTE framework aims to avoid design-related biases, such as selection bias or immortal time bias.3 Subsequently, statistical methods called g-methods are applied within the emulated trial to correct for confounding bias between groups.
Compared with randomised clinical trials (RCTs), TTE studies using observational data can answer a broader set of questions, including comparative effectiveness and safety, comparison of therapeutical strategies or treatment effectiveness in populations usually excluded from RCTs (eg, children, women with a pregnancy plan or older patients). However, the reliability of TTE studies must first be validated, which can be achieved by demonstrating their ability to replicate the results of well-conducted RCTs, the gold standard for causal inference. Such validation studies also provide crucial insights regarding the quality and completeness of the data source, ensuring the feasibility of emulating a target trial and the ability to account for all major confounders.
Recently, the RCT Duplicated Using Prospective Longitudinal Insurance Claims:Applying Techniques of Epidemiology (RCT-DUPLICATE) initiative assessed the concordance and reported substantial agreement between 32 RCT and TTE results using electronic health records in various fields, predominantly on cardiovascular conditions.4 In the field of clinical neurology, TTE could be of major interest, especially for multiple sclerosis (MS) research, which benefits from numerous registries that collect high-quality data generated in clinical routine.5 While statistical methods from the causal inference framework are commonly used in MS research (mainly propensity score matching),6 many observational studies do not attempt to strictly emulate a realistic trial design, which limits the validity of their conclusions. To date, only one study was conducted in the international MSBase registry; it replicated a single trial (TRANSFORMS) focusing solely on clinical outcomes and reported findings consistent with the original RCT.7 However, a broader validation across multiple RCTs and different outcomes is lacking.
In this study, we aimed to evaluate the feasibility and validity of TTE in the MS field by replicating eight RCTs using observational data from the Observatoire Français de la Sclérose en Plaques (OFSEP, the French MS registry), for several clinical and radiological outcomes.
METHODS
RCT selection
Among all RCTs in MS that compared two disease-modifying therapies (DMTs) in the active and control groups (excluding placebo-controlled trials), we selected those evaluating DMTs approved in France before 2021 and commonly used to ensure sufficient patient population in the OFSEP database (ie, excluding alemtuzumab, cladribine, ofatumumab and siponimod). This resulted in the replication of eight RCTs: ASSESS (fingolimod vs glatiramer acetate),8 BEYOND (interferon beta vs glatiramer acetate),9 CONFIRM (dimethyl fumarate (DMF) vs glatiramer acetate),10 OPERA (ocrelizumab vs interferon beta),11 REGARD (interferon beta vs glatiramer acetate),12 RIFUND-MS (rituximab vs DMF),13 TENERE (teriflunomide vs interferon beta)14 and TRANSFORMS (fingolimod vs interferon beta).15
Data source
RCTs were replicated using observational data from the OFSEP database, between 1 January 2003 and 31 December 2022. OFSEP collects data of patients with MS from 42 centres in France.16 Data are reported by neurologists, retrospectively at the first visit and prospectively during patients’ follow-up, using the European Database on Multiple Sclerosis software,17 and comprised demographic (age, sex and children’s birth date), clinical (date of disease onset, relapse occurrences, disability assessed by the Expanded Disability Status Scale (EDSS) score and MS phenotype), radiological (date and results of brain and spinal cord MRIs) and therapeutical (DMT initiation and discontinuation dates) data. Data were extracted on December 8, 2023.
Eligibility criteria
For each emulated trial, patients were included when they initiated one of the DMT evaluated in the corresponding RCT and met its inclusion criteria at the time of DMT initiation regarding recent disease activity (relapse and/or gadolinium-enhancing lesion), age, MS phenotype, EDSS score and disease duration. An additional criterion of calendar year was added to ensure that patients could have received both evaluated DMTs at the time of their inclusion (eg, patients were included in the emulated OPERA trial only if they started their DMT from 2017 onwards, even in the ‘interferon’ group). Exclusion criteria were a recent pregnancy <12 months, a pregnancy during the emulated trial, a DMT discontinuation for pregnancy-related reason during the emulated trial or recent exposure to DMT with a variable delay depending on the DMT (cladribine, alemtuzumab or mitoxantrone at any time before inclusion; anti-CD20 <24 months before; fingolimod or natalizumab <3 months before; methotrexate, azathioprine, mycophenolate mofetil, dimethyl fumarate or teriflunomide <1 month before). Patients could be included in multiple emulated trials but only once per trial, at the first time they met eligibility criteria. Details on the inclusion and exclusion criteria are available in online supplemental table 1.
Treatment assignment and patients’ follow-up
Patients were assigned to the group corresponding to the DMT they initiated in an intention-to-treat setting, regardless of DMT discontinuation or switching after baseline. Time zero was set at the date of DMT initiation, and occurrence of relapses, EDSS measurements as well as brain MRI results were recorded until the end of the emulated trial (1 or 2 years) or at patient’s last follow-up visit, whichever occurred first.
Outcome and causal estimands
The main outcome was the annualised relapse rate (ARR) during the study period. Relapses were reported by the treating neurologist and defined as an acute or subacute episode of patient-reported symptoms and objective findings typical of MS, lasting >24 hours, without fever or infection. The secondary clinical outcome was disability, assessed by the proportion of patients with a 3-month confirmed EDSS progression during the study period. EDSS quantifies disability based on neurological symptoms, clinical examination, walking ability and loss of autonomy; scores range from 0 (normal neurological exam, no symptom) to 10 (death related to MS), in increments of 1 point from 0 to 1, then 0.5 point.18 An EDSS progression was defined by a significant increase compared with the reference EDSS (an increase ≥1.5 for a reference EDSS score of 0, ≥1 for a reference EDSS score from 1 to 5.0, or ≥0.5 for a reference EDSS score ≥5.5), confirmed by a further measurement made >3 months later. This confirmatory EDSS measurement could be recorded during or at any time after the study period. Secondary radiological outcomes were the proportion of patients with ≥1 new or enlarged T2-lesion and the proportion of patients with ≥1 new gadolinium-enhanced T1-lesion on a brain MRI during the study period. Causal estimands of treatment effect were the relative rate (for ARR) or relative risk (for secondary outcomes) between the active and control groups, evaluated as an average treatment effect in the entire study population.
Confounding factors
Confounding factors at baseline were age at inclusion, age at MS onset, sex, calendar year, disease duration, cumulative number of relapses, number of relapses in the past year and in the past 2 years, time between the last relapse and DMT initiation, EDSS score at inclusion, brain and spinal cord MRI lesion-load on the latest available MRI before inclusion, number of MRIs presenting a radiological activity (≥1 new/enlarged T2-lesions and/or new gadolinium-enhanced T1-lesions) in the past year, new gadolinium-enhanced T1-lesion on any MRI in the past year, and number of past DMTs received. For secondary outcomes, the number of EDSS/MRI assessments during follow-up was also accounted for in outcome models.
Censoring and missing outcome assessment
Censoring was addressed by estimating counterfactual treatment effects as if all patients had been followed for the entire study duration. In addition, for secondary outcomes, missing outcome assessments (EDSS measurement or brain MRI) were handled by estimating counterfactual treatment effects as if all patients had sufficient assessments to align with the corresponding RCT protocol. To account for the variability of visit and MRI scheduling in routine practice, we defined broader assessment windows around relevant measurement times. For EDSS progression, patients were required to have at least one EDSS measurement after baseline at 1 year (within 0–1.5 years) for 1-year trials, and two EDSS measurements at 1 year (0–1.25 years) and 2 years (1.25–2.5 years) for 2-year trials. For radiological outcomes, patients were required to have at least one brain MRI at 1 year (within 0.5–1.5 years) for 1-year trials, and at 6 months (0.25–0.75 years), 1 year (0.75–1.5 years) and 2 years (1.5–2.5 years) for 2-year trials. For the analysis of new T2-lesions, brain MRIs had to be compared with an MRI performed at any time after inclusion or within 3 months before. For the analysis of new gadolinium-enhanced T1 lesions, MRI had to be performed with gadolinium injection. Missing-at-random censoring and outcome assessments were accounted for using the same baseline confounders as for the treatment group comparison listed above.
Statistical analysis
Standardised mean differences (SMD) were calculated to assess the balance of baseline covariates between the active and control groups; absolute SMDs above 0.20 indicating large imbalance, between 0.20 and 0.10 moderate imbalance, and below 0.10 good balance.
Treatment effects adjusted for confounding, censoring and missing data were estimated using a targeted maximum likelihood estimator (TMLE), which combines inverse-probability weighting (IPW) and g-computation into a doubly robust estimator.19 IPW was based on two models: an exposure model predicting each patient’s probability of receiving their DMT (ie, the propensity score) and a censoring model to estimate the probability of having a complete follow-up with sufficient outcome assessments. Final weights were computed as the product of the inverse of these two probabilities. G-computation relied on an outcome model, fitted on patients with at least one outcome assessment, which was then used to predict counterfactual outcomes in the entire study population under each DMT, as if all patients had a complete follow-up and sufficient outcome assessments. Finally, the targeting step combined the g-computation estimates (as an offset) and IPW (as weights) in a regression model to improve counterfactual estimations. For each emulated trial, the exposure and censoring models were logistic regressions, and the outcome and targeted models were negative binomial regressions for relapse rates and logistic regressions for secondary outcomes, all adjusted for the confounding factors listed above. Quantitative variables were modelled flexibly using cubic splines or categorised. CIs were obtained by bootstrapping. Analyses were performed with R software, V.4.3.2.20 Details of the statistical methods and models are presented in Supplement.
Concordance between emulated trials and RCTs
Results of the emulated trials were compared with RCTs using three agreement metrics4: regulatory agreement (the effect estimated in the emulated trial has the same direction and statistical significance as the RCT estimate) only for the primary outcome in conclusive RCTs; estimate agreement (the effect estimated in the emulated trial is contained within the 95% CI of the RCT estimate); and standardised difference agreement (the effect estimated in the emulated trial is not statistically different from the RCT estimate using a z-test). Concordance between the emulated trial and RCT was considered achieved when all agreement metrics were met.
RESULTS
A total of 14 111 unique patients were included in the eight emulated trials, ranging from 1411 patients in OPERA to 7781 in REGARD (table 1, flowcharts in online supplemental figure 1).
Table 1. Baseline characteristics of emulated trials and RCTs.
| RCT | Emulated trial | |||
|---|---|---|---|---|
| Active group | Control group | Total | ||
| ASSESS: fingolimod vs glatiramer acetate | n=694 | n=2788 | n=1527 | n=4315 |
| Sex: female, n (%) | 516 (74.4) | 2068 (74.2) | 1245 (81.5) | 3313 (76.8) |
| Age, mean (SD) | 39.9 (10.9) | 38.4 (9.9) | 37.3 (9.8) | 38.0 (9.9) |
| Disease duration, mean (SD) | 7.5 (7.6) | 9.0 (7.3) | 5.4 (6.5) | 7.7 (7.3) |
| EDSS score, mean (SD) | 2.7 (1.4) | 2.2 (1.5) | 1.5 (1.3) | 1.9 (1.5) |
| No prior DMT exposure, n (%) | 324 (46.7) | 654 (23.5) | 1032 (67.6) | 1686 (39.1) |
| Relapse in the last year, mean (SD) | 1.4 (0.8) | 1.4 (0.7) | 1.2 (0.6) | 1.3 (0.7) |
| Relapse in the last 2 years, mean (SD) | 2.2 (1.5) | 1.9 (1.0) | 1.5 (0.8) | 1.8 (1.0) |
| BEYOND: interferon beta vs glatiramer acetate | n=1345 | n=4830 | n=2668 | n=7498 |
| Sex: female, n (%) | 928 (69.0) | 3568 (73.9) | 2121 (79.5) | 5689 (75.9) |
| Age, mean (SD) | 35.5 (NA) | 35.8 (9.0) | 36.9 (8.8) | 36.2 (8.9) |
| Disease duration, mean (SD) | 5.2 (NA) | 5.4 (6.0) | 6.3 (6.5) | 5.7 (6.2) |
| EDSS score, mean (SD) | 2.3 (NA) | 1.6 (1.2) | 1.7 (1.3) | 1.6 (1.3) |
| No prior DMT exposure, n (%) | NA | 3572 (74.0) | 1671 (62.6) | 5243 (69.9) |
| Relapse in the last year, mean (SD) | 1.6 (NA) | 1.4 (0.7) | 1.4 (0.7) | 1.4 (0.7) |
| Relapse in the last 2 years, mean (SD) | NA | 1.8 (1.0) | 1.8 (1.0) | 1.8 (1.0) |
| CONFIRM: DMF vs glatiramer acetate | n=709 | n=2228 | n=957 | n=3185 |
| Sex: female, n (%) | 493 (69.5) | 1608 (72.2) | 785 (82.0) | 2393 (75.1) |
| Age, mean (SD) | 37.2 (9.2) | 36.1 (9.1) | 36.5 (8.9) | 36.2 (9.1) |
| Disease duration, mean (SD) | 7.8 (6.2) | 5.8 (6.4) | 4.9 (6.2) | 5.5 (6.3) |
| EDSS score, mean (SD) | 2.6 (1.2) | 1.4 (1.2) | 1.3 (1.2) | 1.4 (1.2) |
| No prior DMT exposure, n (%) | 507 (71.5) | 1362 (61.1) | 687 (71.8) | 2049 (64.3) |
| Relapse in the last year, mean (SD) | 1.3 (0.6) | 1.2 (0.6) | 1.2 (0.5) | 1.2 (0.6) |
| Relapse in the last 2 years, mean (SD) | NA | 1.5 (0.8) | 1.4 (0.8) | 1.4 (0.8) |
| OPERA: ocrelizumab vs interferon beta | n=1656 | n=969 | n=442 | n=1411 |
| Sex: female, n (%) | 1093 (66.0) | 685 (70.7) | 356 (80.5) | 1041 (73.8) |
| Age, mean (SD) | 37.1 (9.2) | 36.0 (9.0) | 34.0 (9.2) | 35.4 (9.2) |
| Disease duration, mean (SD) | 6.6 (6.1) | 5.8 (6.5) | 2.9 (4.1) | 4.9 (6.0) |
| EDSS score, mean (SD) | 2.8 (1.3) | 2.2 (1.5) | 1.3 (1.1) | 1.9 (1.4) |
| No prior DMT exposure, n (%) | 1209 (73.0) | 474 (48.9) | 343 (77.6) | 817 (57.9) |
| Relapse in the last year, mean (SD) | 1.3 (0.7) | 1.3 (0.6) | 1.1 (0.5) | 1.2 (0.6) |
| Relapse in the last 2 years, mean (SD) | NA | 1.7 (0.9) | 1.4 (0.6) | 1.6 (0.8) |
| REGARD: interferon beta vs glatiramer acetate | n=764 | n=5009 | n=2772 | n=7781 |
| Sex: female, n (%) | 539 (70.5) | 3701 (73.9) | 2203 (79.5) | 5904 (75.9) |
| Age, mean (SD) | 36.7 (9.6) | 36.5 (9.6) | 37.6 (9.4) | 36.9 (9.5) |
| Disease duration, mean (SD) | 6.2 (6.7) | 5.6 (6.2) | 6.5 (6.8) | 5.9 (6.4) |
| EDSS score, mean (SD) | 2.3 (1.3) | 1.6 (1.3) | 1.7 (1.3) | 1.7 (1.3) |
| No prior DMT exposure, n (%) | NA | 3696 (73.8) | 1715 (61.9) | 5411 (69.5) |
| Relapse in the last year, mean (SD) | NA | 1.4 (0.7) | 1.4 (0.7) | 1.4 (0.7) |
| Relapse in the last 2 years, mean (SD) | 1.9 (NA) | 1.8 (1.0) | 1.8 (1.0) | 1.8 (1.0) |
| RIFUND-MS: rituximab vs DMF | n=197 | n=135 | n=1332 | n=1467 |
| Sex: female, n (%) | 131 (66.5) | 93 (68.9) | 944 (70.9) | 1037 (70.7) |
| Age, mean (SD) | 33.4 (7.7) | 33.7 (7.4) | 33.3 (8.2) | 33.4 (8.1) |
| Disease duration, mean (SD) | 1.7 (2.9) | 5.2 (2.9) | 2.7 (2.8) | 2.9 (2.9) |
| EDSS score, mean (SD) | 1.6 (1.1) | 2.3 (1.4) | 1.3 (1.2) | 1.4 (1.2) |
| No prior DMT exposure, n (%) | 190 (96.4) | 33 (24.4) | 961 (72.1) | 994 (67.8) |
| Relapse in the last year, mean (SD) | NA | 1.3 (0.9) | 1.1 (0.6) | 1.1 (0.6) |
| Relapse in the last 2 years, mean (SD) | NA | 1.9 (1.2) | 1.4 (0.7) | 1.5 (0.8) |
| TENERE: teriflunomide vs interferon beta | n=215 | n=1967 | n=1099 | n=3066 |
| Sex: female, n (%) | 148 (68.9) | 1398 (71.1) | 879 (80.0) | 2277 (74.3) |
| Age, mean (SD) | 36.9 (10.4) | 40.9 (11.1) | 36.5 (10.5) | 39.3 (11.1) |
| Disease duration, mean (SD) | 7.1 (7.6) | 7.6 (8.3) | 5.2 (6.7) | 6.7 (7.9) |
| EDSS score, mean (SD) | 2.1 (1.3) | 1.6 (1.3) | 1.4 (1.2) | 1.5 (1.3) |
| No prior DMT exposure, n (%) | 176 (81.8) | 1243 (63.2) | 777 (70.7) | 2020 (65.9) |
| Relapse in the last year, mean (SD) | 1.3 (0.9) | 1.1 (0.5) | 1.2 (0.5) | 1.2 (0.5) |
| Relapse in the last 2 years, mean (SD) | 1.71 | 1.5 (0.7) | 1.5 (0.7) | 1.5 (0.7) |
| TRANSFORMS: fingolimod vs interferon beta | n=866 | n=2456 | n=1883 | n=4339 |
| Sex: female, n (%) | 580 (67.0) | 1814 (73.9) | 1468 (78.0) | 3282 (75.6) |
| Age, mean (SD) | 36.3 (8.5) | 37.3 (8.9) | 35.2 (9.1) | 36.4 (9.0) |
| Disease duration, mean (SD) | 7.4 (6.2) | 8.7 (7.0) | 4.8 (5.8) | 7.0 (6.8) |
| EDSS score, mean (SD) | 2.2 (1.3) | 2.1 (1.4) | 1.4 (1.2) | 1.8 (1.4) |
| No prior DMT exposure, n (%) | 381 (44.0) | 611 (24.9) | 1359 (72.2) | 1970 (45.4) |
| Relapse in the last year, mean (SD) | 1.51 | 1.4 (0.7) | 1.2 (0.6) | 1.3 (0.6) |
| Relapse in the last 2 years, mean (SD) | 2.3 (1.7) | 1.9 (1.0) | 1.6 (0.8) | 1.8 (0.9) |
ASSESS, fingolimod vs glatiramer acetate; BEYOND, interferon beta vs glatiramer acetate; CONFIRM, dimethyl fumarate (DMF) vs glatiramer acetate; DMF, dimethyl fumarate; DMT, disease-modifying therapy; EDSS, expanded disability status scale; NA, not available; OPERA, ocrelizumab vs interferon beta; RCT, randomised clinical trial; REGARD, interferon beta vs glatiramer acetate; RIFUND-MS, rituximab vs DMF; TENERE, teriflunomide vs interferon beta; TRANSFORMS, fingolimod vs interferon beta.
Comparison of RCT and emulated trial populations
Compared with the original RCT population, patients in emulated trials were more frequently female (75.4% in emulated trial vs 68.7% in RCTs) and had lower EDSS (mean (±SD), 1.7 (±1.3) vs 2.5 (±1.3)), but had overall similar age at inclusion, disease duration, number of relapses in the past year and prior exposure to DMT (online supplemental figure 2). DMT discontinuations were more frequent in emulated trials than in RCTs, particularly for moderately effective DMTs (online supplemental table 3). Follow-up completion rates were above 80% and similar to RCTs in most emulated trials, except for the OPERA, TENERE and RIFUND-MS trials in which more patients were lost to follow-up in emulated trials. Sufficient EDSS assessments were available in 55.7%–86.6% of patients, depending on the trials, while the proportion of patients with sufficient MRI assessments was more limited (from 2.0% to 37.7% for new T2-lesions, and 2.9% to 50.2% for new gadolinium-enhanced T1-lesions).
Covariate balance between active and control groups
Before adjustment, differences in confounding factor distribution at baseline between the active and control groups differed among emulated trials (online supplemental table 4 and online supplemental figure 3). Confounders were already well-balanced in the emulated BEYOND, CONFIRM, REGARD and TENERE trials (average absolute SMD of the 15 confounders <0.20), while larger imbalances were observed in the emulated ASSESS, OPERA, RIFUND-MS and TRANSFORMS trials (average absolute SMD ranging from 0.33 to 0.44).
IP-weighted SMD assessed imbalance correction after adjustment by the weighting component of the TMLE (online supplemental figure 4). For the ARR and EDSS progression analyses, this showed good covariate balance after adjustment in most trials, with absolute SMDs reduced below 0.10, although a small imbalance persisted for some covariate in the OPERA trial (between 0.10 and 0.20). Conversely, significant imbalances persisted in the RIFUND trial and for radiological outcome analyses, with some absolute SMD remaining above 0.30. To note, this residual imbalance may still be corrected by the g-computation component of the TMLE.
Relapse rate
Treatment effects on relapse rate estimated in emulated trials were concordant with RCT estimates in 7 of the eight trials after adjustment by TMLE (table 2, figure 1). The exception was the OPERA trial, in which the treatment effect on reduction in relapse risk found in the emulated trial was greater than the RCT estimate (relative relapse rate was 0.20 (95% CI 0.14 to 0.29) in the emulated trial vs 0.53 (95% CI 0.43 to 0.66) in the RCT); this was due to a lower ARR estimated under ocrelizumab (ARR was 0.06 in the emulated trial vs 0.16 in the RCT) with a similar ARR under interferon (ARR was 0.28 vs 0.29). In contrast, absolute ARRs in the active and control groups estimated in emulated trials were higher than in RCTs in most cases; the mean ARR in active groups was 0.27 relapse/year in emulated trials versus 0.20 in RCTs and the mean ARR in control groups was 0.36 in emulated trials versus 0.26 in RCTs (online supplemental figure 5). For three trials, the unadjusted effects estimated in emulated trials were not concordant with RCT estimates (OPERA, RIFUND-MS and TRANSFORMS) and became concordant after adjustment for baseline confounding factors in the last two trials. In the other trials, the unadjusted effect estimates were already concordant with RCT estimates but moved closer after adjustment for three trials (ASSESS, CONFIRM and TENERE, figure 2).
Table 2. Effect estimates in RCTs and emulated trials for the annualised relapse rate.
| Trial | RCT | Emulated trial | ||||
|---|---|---|---|---|---|---|
| ARR | Relative rate (95% CI) | ARR | Relative rate (95% CI) | |||
| Active | Control | Active | Control | |||
| ASSESS | 0.15 | 0.26 | 0.59 (0.37 to 0.95) | 0.29 | 0.46 | 0.62 (0.53 to 0.72)* |
| BEYOND | 0.36 | 0.34 | 1.06 (0.89 to 1.26) | 0.44 | 0.44 | 1.00 (0.94 to 1.07)* |
| CONFIRM | 0.22 | 0.29 | 0.76 (0.56 to 1.03) | 0.22 | 0.27 | 0.79 (0.69 to 0.90)* |
| OPERA | 0.16 | 0.29 | 0.53 (0.43 to 0.66) | 0.06 | 0.28 | 0.20 (0.14 to 0.29)† |
| REGARD | 0.30 | 0.29 | 1.03 (0.85 to 1.25) | 0.44 | 0.44 | 1.00 (0.93 to 1.07)* |
| RIFUND-MS | 0.01 | 0.09 | 0.19 (0.06 to 0.62) | 0.06 | 0.22 | 0.25 (0.15 to 0.57)* |
| TENERE | 0.26 | 0.22 | 1.20 (0.62 to 2.30) | 0.27 | 0.28 | 0.95 (0.83 to 1.07)* |
| TRANSFORMS | 0.16 | 0.33 | 0.48 (0.34 to 0.70) | 0.29 | 0.49 | 0.60 (0.53 to 0.69)* |
Satisfied all agreement metrics: regulatory agreement: the adjusted effect estimated in the emulated trial has the same direction and statistical significance as the RCT estimate (only for conclusive RCTs); EA: the adjusted effect estimated in the emulated trial is contained within the 95% CI of the RCT estimate; standardised difference agreement: the adjusted effect estimated in the emulated trial is not statistically different from the RCT estimate using a z-test.
Satisfied only the regulatory agreement metric.
ARR, annualised relapse rate; ASSESS, fingolimod vs glatiramer acetate; BEYOND, interferon beta vs glatiramer acetate; CONFIRM, DMF vs glatiramer acetate; DMF, dimethyl fumarate; EA, estimate agreement; OPERA, ocrelizumab vs interferon beta; RA, regulatory agreement; RCT, randomised clinical trial; REGARD, interferon beta vs glatiramer acetate; RIFUND-MS, rituximab vs DMF; TENERE, teriflunomide vs interferon beta; TRANSFORMS, fingolimod vs. interferon beta .
Figure 1. Concordance between the treatment effect estimated in emulated trials and RCTs according to outcome EA: the adjusted effect estimated in the emulated trial is contained within the 95% CI of the RCT estimate; standardised difference agreement: the adjusted effect estimated in the emulated trial is not statistically different from the RCT estimate using a z-test. ASSESS, fingolimod vs glatiramer acetate; BEYOND, interferon beta vs glatiramer acetate; CONFIRM, DMF vs glatiramer acetate; DMF, dimethyl fumarate; EA, estimate agreement; EDSS, expanded disability status scale; OPERA, ocrelizumab vs interferon beta; RCT, randomised clinical trial; REGARD, interferon beta vs glatiramer acetate; RIFUND-MS, rituximab vs DMF; SDA, standardised difference agreement; TENEREvs interferon beta; TRANSFORMSvs interferon beta.
Figure 2. Treatment effect estimated in emulated trials and RCTs according to outcome, before and after adjustment for confounding factors. Coloured boxes indicate RCT estimates (estimate and 95% CI) and are coloured according to the concordance between emulated trial and RCT estimates. ARR, annualised relapse rate; ASSESS, fingolimod vs glatiramer acetate; BEYOND, interferon beta vs glatiramer acetate; CONFIRM, DMF vs glatiramer acetate; EA, estimate agreement; EDSS, expanded disability status scale; OPERA, ocrelizumab vs interferon beta; RCT, randomised clinical trial; REGARD, interferon beta vs glatiramer acetate; RIFUND-MS, rituximab vs DMF; SDA, standardised difference agreement; TENERE, teriflunomide vs interferon beta; TMLE, targeted maximum likelihood estimator; TRANSFORMS, fingolimod vs interferon beta.
EDSS progression
Treatment effects on the risk of EDSS progression estimated in emulated trials were concordant with RCT estimates after adjustment in the six trials that reported this outcome (table 3). The absolute proportion of patients with an EDSS progression during the study period in the emulated trials was also higher than in RCTs; the average proportion of patients with EDSS progression in active groups was 21.3% in emulated trials versus 11.8% in RCTs, while in control groups, it was 19.7% versus 11.9%, respectively. The unadjusted effect estimate was not concordant with the RCT estimate for the OPERA trial but was already concordant for other trials.
Table 3. Effect estimates in RCTs and emulated trials for secondary outcomes.
| Trial | RCT | Emulated trial | |||||
|---|---|---|---|---|---|---|---|
| Proportion of patients with outcome | Relative risk (95% CI) | n with sufficient EDSS/MRI assessment | Proportion of patients with outcome | Relative risk (95% CI) | |||
| Active | Control | Active | Control | ||||
| EDSS progression | |||||||
| ASSESS | – | – | – | 3564 | 0.16 | 0.16 | 1.02 (0.85 to 1.25) |
| BEYOND | 0.21 | 0.20 | 1.05 (0.82 to 1.34) | 4028 | 0.27 | 0.24 | 1.11 (0.98 to 1.24)* |
| CONFIRM | 0.13 | 0.16 | 0.81 (0.53 to 1.24) | 1904 | 0.19 | 0.21 | 0.90 (0.73 to 1.12)* |
| OPERA | 0.09 | 0.14 | 0.67 (0.50 to 0.90) | 803 | 0.19 | 0.27 | 0.72 (0.51 to 1.16)* |
| REGARD | 0.12 | 0.09 | 1.34 (0.84 to 2.15) | 4174 | 0.27 | 0.24 | 1.12 (1.01 to 1.26)* |
| RIFUND-MS | 0.10 | 0.05 | 2.00 (0.70 to 5.72) | 866 | 0.30 | 0.16 | 1.86 (0.96 to 2.97)* |
| TENERE | – | – | – | 1858 | 0.25 | 0.27 | 0.94 (0.79 to 1.14) |
| TRANSFORMS | 0.06 | 0.08 | 0.75 (0.44 to 1.26) | 3559 | 0.16 | 0.16 | 1.05 (0.87 to 1.32)* |
| New/enlarged T2-lesions | |||||||
| ASSESS | 0.48 | 0.65 | 0.75 (0.65 to 0.87) | 1436 | 0.27 | 0.37 | 0.72 (0.59 to 0.89)* |
| BEYOND | – | – | – | 192 | 0.56 | 0.63 | 0.88 (0.67 to 1.15) |
| CONFIRM | 0.73 | 0.76 | 0.96 (0.84 to 1.10) | 226 | 0.57 | 0.73 | 0.78 (0.63 to 0.99)† |
| OPERA | 0.37 | 0.61 | 0.61 (0.55 to 0.68) | 104 | 0.29 | 0.69 | 0.42 (0.24 to 0.62)† |
| REGARD | 0.60 | 0.63 | 0.95 (0.82 to 1.10) | 195 | 0.55 | 0.62 | 0.89 (0.67 to 1.14)* |
| RIFUND-MS | 0.21 | 0.37 | 0.58 (0.36 to 0.91) | 122 | NA ‡ | NA ‡ | NA ‡ |
| TENERE | – | – | – | 186 | 0.61 | 0.55 | 1.10 (0.88 to 1.48) |
| TRANSFORMS | 0.45 | 0.54 | 0.83 (0.72 to 0.96) | 1342 | 0.28 | 0.38 | 0.75 (0.60 to 0.92)* |
| New gadolinium-enhanced T1-lesion | |||||||
| ASSESS | 0.14 | 0.23 | 0.59 (0.41 to 0.84) | 2075 | 0.22 | 0.32 | 0.68 (0.57 to 0.86)* |
| BEYOND | – | – | – | 259 | 0.56 | 0.69 | 0.81 (0.67 to 0.97) |
| CONFIRM | – | – | – | 229 | 0.38 | 0.53 | 0.73 (0.53 to 0.99) |
| OPERA | 0.05 | 0.30 | 0.17 (0.12 to 0.23) | 96 | 0.06 | 0.63 | 0.09 (0.01 to 0.22)† |
| REGARD | 0.19 | 0.33 | 0.58 (0.42 to 0.80) | 265 | 0.55 | 0.67 | 0.81 (0.66 to 0.96)† |
| RIFUND-MS | – | – | – | 122 | NA ‡ | NA ‡ | NA ‡ |
| TENERE | – | – | – | 207 | 0.44 | 0.42 | 1.06 (0.75 to 1.37) |
| TRANSFORMS | 0.10 | 0.19 | 0.52 (0.36 to 0.75) | 2052 | 0.23 | 0.25 | 0.93 (0.75 to 1.12) |
Satisfied all agreement metrics. EA: the adjusted effect estimated in the emulated trial is contained within the 95% CI of the RCT estimate; standardised difference agreement: the adjusted effect estimated in the emulated trial is not statistically different from the RCT estimate using a z-test.
Satisfied only the estimate agreement metric.
The number of patients with an available MRI assessment was not sufficient in the active group (n=10) to estimate the effect of treatment on radiological outcomes in the emulation of the RIFUND-MS trial.
ASSESS, fingolimod vs glatiramer acetate; BEYOND, interferon beta vs glatiramer acetate; CONFIRM, DMF vs glatiramer acetate; DMF, dimethyl fumarate; EA, estimate agreement; EDSS, expanded disability status scale; NA, not available; OPERA, ocrelizumab vs interferon beta; RCT, randomised clinical trial; REGARD, interferon beta vs glatiramer acetate; RIFUND-MS, rituximab vs DMF; TENERE, teriflunomide vs interferon beta; TRANSFORMS, fingolimod vs. interferon beta.
Radiological outcomes
Among the five trials in which the appearance of new T2-lesions was evaluated, relative risks of new/enlarged T2-lesions estimated in emulated trials were concordant with RCT estimates in three trials (ASSESS, REGARD and TRANSFORMS), whereas only the standardised difference agreement criterion was met for CONFIRM and OPERA trials. Among the four trials in which the proportion of patients with a new gadolinium-enhanced T1-lesion was reported, relative risks of new gadolinium-enhanced T1-lesions estimated in emulated trials were concordant with RCT estimates in one trial (ASSESS). In two trials (OPERA and REGARD), only the standardised difference agreement was met, while in one trial (TRANSFORMS), it was significantly different from the RCT estimate. For the RIFUND-MS trial, the treatment effect on radiological outcome could not be estimated in the emulated trial due to the very small number of patients with sufficient MRI assessments in the ‘rituximab’ group (n=10).
DISCUSSION
In the present study, we successfully replicated eight RCTs evaluating the efficacy of DMTs in MS using a TTE approach applied to observational data from the French MS registry. The findings demonstrated a strong concordance between emulated trial and RCT estimates for two clinical outcomes, relapse rate and EDSS progression, except in the OPERA trial (ocrelizumab vs interferon) for the analysis of relapse rate. Replication of radiological outcomes was more challenging; fewer emulated trials achieved concordance with RCTs (three out of five trials for new/enlarged T2-lesions, and one out of four for new gadolinium-enhanced T1-lesions). In contrast to the good concordance for relative treatment effect estimations, the absolute outcome values in our emulated trials differed substantially from those of RCTs. Specifically, ARRs and proportions of patients with EDSS progression were generally higher in both the active and placebo groups of emulated trials than in RCTs.
Considering the increasing emphasis on real-world evidence in regulatory decision-making,21 validation studies comparing results of RCTs and TTE are essential. The RCT-DUPLICATE study demonstrated good concordance between 32 emulated trials and their corresponding RCT: 66% achieved estimate agreement and 75% standardised difference agreement.4 Of note, the emulated trials were constructed using electronic health records rather than registry data, and none was in the neurological field. In MS, the prior study that replicated the TRANSFORMS trial using a TTE approach and the international MSBase registry also found a good concordance with the corresponding RCT.7 This study evaluated clinical outcomes, relapse rate and EDSS, using propensity score matching, but validation studies should encompass multiple trials simultaneously. In addition to this study, our findings provide strong support for the credibility of real-world evidence based on TTE using registry data to address causal questions in MS.
Comparison between emulated trials and RCTs is challenging due to many potential sources of discrepancy.22 First, residual confounding remains the main concern in observational studies as it compromises confidence in their results. The good concordance in the present study suggests that adjusting for the set of confounders we selected is adequate to estimate unbiased treatment effects using observational data in MS. While unadjusted estimates were already concordant in some cases, adjustment for confounding factors proved essential for achieving good concordance, particularly when baseline characteristics between the two groups were dissimilar. However, each study question may have specific confounders, depending on the intervention and outcome being evaluated, and also potentially the data source. Furthermore, our findings underline that sufficient sample size is critical for covariate adjustment, as shown in settings with fewer patients like the RIFUND trial and for radiological outcomes.
Second, even when inclusion and exclusion criteria are correctly replicated, populations might differ between RCTs and observational studies, due to the selection process inherent to RCTs and variations in physician prescribing behaviour, patient preferences, treatment accessibility or availability of alternative therapeutic strategies. Moreover, inconsistencies between the treatment strategy evaluated in the RCT and its application in practice may occur, including differences in treatment adherence, discontinuation or switching. As an example, for the OPERA trial, DMT discontinuations or switches were more frequent in the interferon group in our emulated trial (~50%) than in the RCT (~15%), and less frequent in the ocrelizumab group (~5% vs ~10%).
Finally, differences in outcome assessments between RCT and real-life practice must also be considered. The analysis of radiological outcomes was primarily complicated by the stricter MRI protocols in RCTs, particularly regarding timing of MRI assessment and gadolinium injections. Variability in MRI technical parameters, interpretation and radiological outcome definitions can also exist not only between RCTs and real-world practice but also among various RCT protocols.
Importantly, discordance between emulated trial and RCT results does not necessarily indicate bias in the emulated trial estimate. Aside from residual confounding, the other sources of difference mentioned above may reflect the limited transposability of RCT results to real-life practice. In fact, the higher-than-expected effectiveness of ocrelizumab found herein was consistent across clinical and radiological outcomes and aligned with clinical practice expectations and other observational studies.23,26
Our study has several limitations. First, we only replicated trials with active comparators, as placebo-controlled designs pose greater challenges in defining an appropriate time zero and assigning a treatment strategy for untreated patients. Although these issues can be addressed using cloning/censoring/weighting,27 it considerably increases methodological complexity, requiring longitudinal modelling. Second, certain aspects of the study protocols inevitably diverged between RCTs and emulated trials, such as exclusion criteria, particularly regarding comorbidities or past DMT exposure, but these differences probably had minimal impact. Finally, validating TTE through the replication of RCTs is only one step towards the broader acceptance of real-world evidence in MS but does not guarantee that all observational studies using a TTE methodology will yield unbiased causal estimates. Their reliability will depend on multiple factors, including quality of the data source, appropriate causal reasoning for confounders selection (always with a risk of residual confounding) and key methodological decisions such as the definition of the causal estimand, TTE protocol, adjustment method and model construction.
Conclusion
The combined use of a TTE methodology and high-quality registry data is a valid and powerful tool to evaluate treatment effectiveness in MS, particularly for relapse rate and disability progression. Our findings support the use of observational data from MS registries to explore questions beyond the scope of RCTs, such as comparative effectiveness, treatment effects in specific populations and comparisons of therapeutic strategies.
Supplementary material
Acknowledgements
Data collection has been supported by a grant provided by the French State and handled by the Agence Nationale de la Recherche, within the framework of the France 2030 program, under the reference ANR-10-COHO-002, Observatoire Français de la Sclérose en Plaques (OFSEP) and by the Eugène Devic EDMUS Foundation against multiple sclerosis.The authors thank Shanez Haouari (Direction de la Recherche en Santé, Hospices Civils de Lyon) for help in manuscript preparation.
Footnotes
Funding: The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.
Provenance and peer review: Not commissioned; externally peer-reviewed.
Patient consent for publication: Not applicable.
Ethics approval: OFSEP (clinicaltrials.gov [NCT02889965]) was approved by the French data protection agency (authorisation request 914066v3) and an institutional review board (reference 2019-A03066-51). Participants gave informed consent to participate in the study before taking part.
Data availability free text: Data may be made available at the motivated request by any expert researcher, after approval by the OFSEP.
Collaborators: For the OFSEP investigators.
Data availability statement
Data may be obtained from a third party and are not publicly available.
References
- 1.Hernán MA, Wang W, Leaf DE. Target trial emulation: a framework for causal inference from observational data. JAMA. 2022;328:2446–7. doi: 10.1001/jama.2022.21383. [DOI] [PubMed] [Google Scholar]
- 2.Desai RJ, Wang SV, Sreedhara SK, et al. Process guide for inferential studies using healthcare data from routine clinical practice to evaluate causal effects of drugs (PRINCIPLED): considerations from the FDA Sentinel Innovation Center. BMJ. 2024;384 doi: 10.1136/bmj-2023-076460. [DOI] [PubMed] [Google Scholar]
- 3.Hernán MA, Sauer BC, Hernández-Díaz S, et al. Specifying a target trial prevents immortal time bias and other self-inflicted injuries in observational analyses. J Clin Epidemiol. 2016;79:70–5. doi: 10.1016/j.jclinepi.2016.04.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Wang SV, Schneeweiss S, RCT-DUPLICATE Initiative. et al. Emulation of randomized clinical trials with nonrandomized database analyses: results of 32 clinical trials. JAMA. 2023;329:1376–85. doi: 10.1001/jama.2023.4221. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Trojano M, Kalincik T, Iaffaldano P, et al. Interrogating large multiple sclerosis registries and databases: what information can be gained? Curr Opin Neurol. 2022;35:271–7. doi: 10.1097/WCO.0000000000001057. [DOI] [PubMed] [Google Scholar]
- 6.Karim ME, Pellegrini F, Platt RW, et al. The use and quality of reporting of propensity score methods in multiple sclerosis literature: A review. Mult Scler . 2022;28:1317–23. doi: 10.1177/1352458520972557. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Signori A, Ponzano M, Kalincik T, et al. Emulating randomised clinical trials in relapsing-remitting multiple sclerosis with non-randomised real-world evidence: an application using data from the MSBase Registry. J Neurol Neurosurg Psychiatry . 2024;95:620–5. doi: 10.1136/jnnp-2023-332603. [DOI] [PubMed] [Google Scholar]
- 8.Cree BAC, Goldman MD, Corboy JR, et al. Efficacy and safety of 2 fingolimod doses vs glatiramer acetate for the treatment of patients with relapsing-remitting multiple sclerosis. JAMA Neurol. 2021;78:48. doi: 10.1001/jamaneurol.2020.2950. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.O’Connor P, Filippi M, Arnason B, et al. 250 microg or 500 microg interferon beta-1b versus 20 mg glatiramer acetate in relapsing-remitting multiple sclerosis: a prospective, randomised, multicentre study. Lancet Neurol. 2009;8:889–97. doi: 10.1016/S1474-4422(09)70226-1. [DOI] [PubMed] [Google Scholar]
- 10.Fox RJ, Miller DH, Phillips JT, et al. Placebo-controlled phase 3 study of oral BG-12 or glatiramer in multiple sclerosis. N Engl J Med. 2012;367:1087–97. doi: 10.1056/NEJMoa1206328. [DOI] [PubMed] [Google Scholar]
- 11.Hauser SL, Bar-Or A, Comi G, et al. Ocrelizumab versus Interferon Beta-1a in Relapsing Multiple Sclerosis. N Engl J Med. 2017;376:221–34. doi: 10.1056/NEJMoa1601277. [DOI] [PubMed] [Google Scholar]
- 12.Mikol DD, Barkhof F, Chang P, et al. Comparison of subcutaneous interferon beta-1a with glatiramer acetate in patients with relapsing multiple sclerosis (the REbif vs Glatiramer Acetate in Relapsing MS Disease [REGARD] study): a multicentre, randomised, parallel, open-label trial. Lancet Neurol. 2008;7:903–14. doi: 10.1016/S1474-4422(08)70200-X. [DOI] [PubMed] [Google Scholar]
- 13.Svenningsson A, Frisell T, Burman J, et al. Safety and efficacy of rituximab versus dimethyl fumarate in patients with relapsing-remitting multiple sclerosis or clinically isolated syndrome in Sweden: a rater-blinded, phase 3, randomised controlled trial. Lancet Neurol. 2022;21:693–703. doi: 10.1016/S1474-4422(22)00209-5. [DOI] [PubMed] [Google Scholar]
- 14.Vermersch P, Czlonkowska A, Grimaldi LM, et al. Teriflunomide versus subcutaneous interferon beta-1a in patients with relapsing multiple sclerosis: a randomised, controlled phase 3 trial. Mult Scler . 2014;20:705–16. doi: 10.1177/1352458513507821. [DOI] [PubMed] [Google Scholar]
- 15.Cohen JA, Barkhof F, Comi G, et al. Oral fingolimod or intramuscular interferon for relapsing multiple sclerosis. N Engl J Med. 2010;362:402–15. doi: 10.1056/NEJMoa0907839. [DOI] [PubMed] [Google Scholar]
- 16.Vukusic S, Casey R, Rollot F, et al. Observatoire Français de la Sclérose en Plaques (OFSEP): a unique multimodal nationwide MS registry in France. Mult Scler J. 5 doi: 10.1177/1352458518815602. n.d. [DOI] [PubMed] [Google Scholar]
- 17.Confavreux C, Compston DA, Hommes OR, et al. EDMUS, a European database for multiple sclerosis. J Neurol Neurosurg Psychiatry . 1992;55:671–6. doi: 10.1136/jnnp.55.8.671. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Kurtzke JF. Rating neurologic impairment in multiple sclerosis: an expanded disability status scale (EDSS) Neurology (ECronicon) 1983;33:1444–52. doi: 10.1212/wnl.33.11.1444. [DOI] [PubMed] [Google Scholar]
- 19.van der Laan MJ, Rubin D. Targeted Maximum Likelihood Learning. Int J Biostat. 2006;2 doi: 10.2202/1557-4679.1043. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.R: A language and environment for statistical computing . Vienna, Austria: R Core Team; 2020. R foundation for statistical computing [internet]https://www.R-project.org/ Available. [Google Scholar]
- 21.US Food and Drug Administration . Framework for FDA’s Real-World Evidence Program. Washington, DC: US FDA; 2018. [Google Scholar]
- 22.Franklin JM, Glynn RJ, Suissa S, et al. Emulation differences vs. biases when calibrating real-world evidence findings against randomized controlled trials. Clin Pharmacol Ther. 2020;107:735–7. doi: 10.1002/cpt.1793. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Sempere AP, Berenguer-Ruiz L, Borrego-Soriano I, et al. Ocrelizumab in multiple sclerosis: a real-world study from Spain. Front Neurol. 2020;11 doi: 10.3389/fneur.2020.592304. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Smoot K, Chen C, Stuchiner T, et al. Clinical outcomes of patients with multiple sclerosis treated with ocrelizumab in a US community MS center: an observational study. BMJ Neurol Open. 2021;3 doi: 10.1136/bmjno-2020-000108. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Pontieri L, Blinkenberg M, Bramow S, et al. Ocrelizumab treatment in multiple sclerosis: a Danish population-based cohort study. Eur J Neurol. 2022;29:496–504. doi: 10.1111/ene.15142. [DOI] [PubMed] [Google Scholar]
- 26.Foong YC, Merlo D, Gresle M, et al. Comparing ocrelizumab to interferon/glatiramer acetate in people with multiple sclerosis over age 60. J Neurol Neurosurg Psychiatry . 2024;95:767–74. doi: 10.1136/jnnp-2023-332883. [DOI] [PubMed] [Google Scholar]
- 27.Hernán MA, Robins JM. Using big data to emulate a target trial when a randomized trial is not available. Am J Epidemiol. 2016;183:758–64. doi: 10.1093/aje/kwv254. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Data may be obtained from a third party and are not publicly available.