Elamipretide MMPOWER-3 Phase 3 Trial: NCT03323749

MMPOWER-3 was a 24-week, randomized, double-blind, placebo-controlled phase 3 trial in 218 participants with genetically confirmed primary mitochondrial myopathy. It did not meet either co-primary endpoint: six-minute walk distance or the Primary Mitochondrial Myopathy Symptom Assessment total fatigue score. The study is a direct example of why biological rationale, earlier signals, and later exploratory subgroups cannot replace prespecified phase 3 outcomes.[1][3]

FDA later granted accelerated approval to the finished product Forzinity (elamipretide) to improve muscle strength in adult and pediatric patients with Barth syndrome weighing at least 30 kg. That different-disease, product-specific indication remains subject to confirmation of clinical benefit and does not change the null MMPOWER-3 findings in primary mitochondrial myopathy.[4][5][6][9]

Study at a glance

The intervention description identifies the experiment only. It is not an administration recommendation.

The prespecified endpoint hierarchy

The two primary endpoints were the week-24 changes in:

1. distance walked during the six-minute walk test; and
2. total fatigue on the PMMSA, a patient-reported scale in which a lower score represents less fatigue.

The registered statistical analysis plan used Hochberg's procedure to control the primary family's false-positive error rate at 5%. If both primary p-values were at most 0.05 and favored treatment, both could be declared significant; otherwise, the smaller p-value had to be at most 0.025. Both primary endpoints had to succeed before confirmatory testing could proceed through the ordered secondary outcomes. The plan also specified reporting nominal p-values regardless of this gate.[1][8]

Neither primary endpoint succeeded. Later secondary or subgroup findings therefore cannot be presented as if the phase 3 trial succeeded. The published 95% confidence intervals below are not described here as multiplicity-adjusted intervals.

Six-minute walk result

In the detailed Results section, the least-squares mean change at week 24 was +14.1 meters with elamipretide and +17.3 meters with placebo. The between-group difference was −3.2 meters, with a 95% confidence interval from −18.7 to +12.3 and p=0.69.[1]

The point estimate did not favor elamipretide, and the interval includes both better and worse walking performance relative to placebo. The per-protocol analysis was consistent with the primary analysis: a −2.2-meter difference, 95% CI −16.9 to +12.5, p=0.77.[1]

Fatigue result and a reporting discrepancy

The detailed Results section and Figure 2B report least-squares mean PMMSA fatigue changes of −1.13 with elamipretide and −1.05 with placebo. Its between-group difference was −0.07 points, with a 95% confidence interval from −0.69 to +0.54 and p=0.81.[1]

The article abstract reports the same −0.07 point estimate but a different confidence interval and p-value: 95% CI −0.10 to +0.26 and p=0.37.[1][2] ClinicalTrials.gov posts group means but no between-group analysis for this outcome.[3] The published paper's detailed Results and Figure 2B agree with each other, so their figures are used here while the unresolved abstract-versus-detailed-results conflict is disclosed. The posted registry group means are not the model-adjusted least-squares means quoted above and do not resolve this disagreement.

Under either reported set of statistics, the fatigue endpoint was not met. This agreement about the null conclusion does not make the conflicting numerical reports interchangeable.

Secondary and exploratory findings

The paper states that the study did not meet its secondary endpoints. It also describes exploratory genotype analyses after pooling a heterogeneous set of mitochondrial and nuclear DNA disorders. A signal in the nuclear-DNA subgroup was labeled hypothesis-generating, and three participants' genotype classifications were changed during post-hoc analysis.[1]

Those analyses may guide a future targeted trial. They do not overturn the null overall population result, and they should not be read as replicated evidence for a genotype-defined benefit.

Adverse events

During the double-blind period, adverse events were reported by 107 of 109 elamipretide participants (98.2%) and 83 of 109 placebo participants (76.1%). The most common events with elamipretide were injection-site reactions. Serious adverse events occurred in 5 of 109 participants (4.6%) with elamipretide and 3 of 109 (2.8%) with placebo; the paper states they were not considered treatment-related. The paper's safety table reports adverse events leading to treatment discontinuation in 8 of 109 (7.3%) and 2 of 109 (1.8%), respectively. These are not the smaller adverse-event counts for leaving the study altogether. No deaths were reported during the double-blind study.[1][3]

“Generally well tolerated” does not mean risk-free. The high rate of local reactions, the short controlled duration, and the limited sample remain part of the safety interpretation. Our guide to adverse events and tolerability explains why seriousness, treatment relatedness, and treatment discontinuation are separate fields.

Attrition, missing data, and the extension

Of 296 people screened, 218 were randomized and received study treatment. A total of 205 completed the double-blind study period: 102 of 109 with elamipretide and 103 of 109 with placebo. Thirteen did not complete that study period, which is different from discontinuing treatment. The registry lists nine lost to follow-up (four active, five placebo) and four adverse-event study withdrawals (three active, one placebo). The paper's flow diagram places the nine in a broader patient-decision category that includes loss to follow-up.[1][3]

Week-24 measurements were available for 100 active and 102 placebo participants for the walk test, and 98 and 101 for fatigue. Those observed-visit denominators are not the full randomized population.[1][3]

The protocol and statistical analysis plan define the population, endpoints, and analysis rather than a universal effect of the molecule. The plan's study-defined intention-to-treat set included participants who received at least one dose, analyzed by randomized assignment. It explicitly included data collected after treatment discontinuation but before withdrawal from the study. Accordingly, the analysis should not be described as excluding all post-treatment-discontinuation observations or as estimating only a hypothetical uninterrupted-treatment effect.[7][8]

The primary mixed-model repeated-measures analysis used available scheduled observations without a separate primary imputation step. The plan specified multiple-imputation sensitivity analysis when more than 5% of an endpoint's data were missing in either group, a per-protocol sensitivity analysis, and a conditional tipping-point analysis if the multiple-imputation result was significant. These are planned methods; this summary does not claim that every planned sensitivity analysis was reported or independently reproduced. Missing-data assumptions still matter even with similar completion rates.[8] See the guide to trial estimands and missing data for the distinction between the treatment question and unavailable measurements.

The open-label extension enrolled 196 participants: 93 from the prior active group and 103 from the prior placebo group. None completed the planned extension. The registry assigns 171 discontinuations to termination of the clinical-development program (83 and 88), 13 to adverse events (three and ten), and 12 to participant withdrawal (seven and five). These reasons sum to 196; neither all 196 administrative terminations nor all 196 adverse-event withdrawals would be accurate. Both extension groups received elamipretide, the registry's prior-placebo grouping does not denote an ongoing placebo comparison.[3]

Sponsor role and study interpretation

Stealth BioTherapeutics funded the trial and participated in study-design development. The paper states that all authors participated in data collection, interpretation, and writing the clinical study report. The lead author had full access to the data; the other authors received aggregate analyses. All authors made the final decision to submit the manuscript.[1]

Everest Clinical Research performed statistical analyses, and Jeffrey Finman of Jupiter Point Pharma Consulting performed post-hoc analyses. Writing and editing assistance included James Shiffer of Write On Time and David Brown and Gene Kelly of Stealth; Stealth personnel also supported endpoint development and statistical interpretation. These disclosed roles are part of the evidence record, not evidence of independent review by this website.[1]

These disclosures do not erase the randomized design or the null result. They show why protocol access, registry comparison, endpoint hierarchy, and transparent reporting remain essential.

Statistical significance, effect size, and clinical importance

• Statistical significance: neither co-primary comparison met its prespecified criterion; the reported p-values in the detailed Results section were 0.69 and 0.81.
• Effect size: the between-group estimates were −3.2 meters on the walk test and −0.07 points on the fatigue scale, not merely “negative” labels.
• Uncertainty: both detailed-results confidence intervals crossed zero and allowed better or worse outcomes; the fatigue interval also conflicts with the abstract despite verification against the published paper.
• Clinical importance: the tested outcomes address walking capacity and fatigue, but no favorable statistically stable effect was demonstrated in the overall population.

Important limitations

• MMPOWER-3 pooled genetically heterogeneous mitochondrial and nuclear DNA disorders in one basket population.
• Participants were mostly White (94%) and had mild-to-moderate baseline impairment on several measures, limiting some generalizability and sensitivity questions.
• Both primary outcomes depended partly on effort or patient report, and the paper discusses placebo response and the absence of validated objective biomarkers.
• The controlled period was 24 weeks; the planned long-term extension was terminated.
• Exploratory genotype findings involved small subgroups and post-hoc reclassification.
• The article's abstract and detailed Results section conflict on the fatigue endpoint's confidence interval and p-value.
• The sponsor funded the trial and participated in design and interpretation.

What this study does not prove

MMPOWER-3 does not show that elamipretide improved walking distance or fatigue in the overall primary-mitochondrial-myopathy population. It also does not prove that the true effect is exactly zero or establish equivalence; the confidence intervals describe a range of effects compatible with the data and model. The guide to negative and null study results explains this distinction.

The trial does not establish outcomes for Barth syndrome. Conversely, FDA's September 19, 2025 accelerated approval of Forzinity concerns improvement in muscle strength in adult and pediatric patients with Barth syndrome weighing at least 30 kg. It was based on knee-extensor muscle strength as an intermediate clinical endpoint; continued approval may depend on confirmation of clinical benefit. FDA's current ongoing-approvals table still lists the confirmatory requirement. This finished-product-specific evidence for a different disease and endpoint cannot be imported into MMPOWER-3.[4][5][6][9]

No finding here authenticates or validates an independently sourced research material described as elamipretide, SS-31, MTP-131, or Bendavia. An approved drug and a research material are not interchangeable merely because they share a name.

Frequently asked questions

Did MMPOWER-3 meet either primary endpoint?

No. Neither six-minute walk distance nor PMMSA total fatigue differed significantly from placebo at week 24.

What was the six-minute walk effect estimate?

The detailed Results section reports a −3.2-meter difference, 95% CI −18.7 to +12.3, p=0.69.

Why are two fatigue statistics mentioned?

The abstract and detailed Results section report different intervals and p-values for the same −0.07 point estimate. Both are null, and the discrepancy remains unresolved in the cited records.

Did a genotype subgroup make the trial positive?

No. The nuclear-DNA subgroup signal was exploratory and hypothesis-generating. It cannot replace the null overall primary outcomes.

Why is the registry marked terminated?

The 24-week double-blind period finished, but the planned open-label extension was ended after the primary endpoints were not met.

Is elamipretide FDA approved?

As checked September 2, 2026 in Edmonton (September 3 UTC), FDA's current labeling and ongoing-approvals page support accelerated approval of Forzinity to improve muscle strength in adult and pediatric patients with Barth syndrome weighing at least 30 kg, with a confirmatory requirement. That is not approval for primary mitochondrial myopathy or for unrelated materials.[6][9]

Does Forzinity approval change MMPOWER-3's result?

No. Different disease, product-specific evidence, and different endpoints cannot rewrite this trial's null primary outcomes.

Primary and official sources

1. Primary paper: MMPOWER-3 in Neurology; published open-access PDF in NLM's PMC collection
2. PubMed record: MMPOWER-3, PMID 37268435
3. ClinicalTrials.gov: NCT03323749 registration and posted results
4. FDA: Accelerated approval of Forzinity for Barth syndrome
5. FDA approval letter: Forzinity, NDA 215244
6. FDA: Ongoing non-oncology accelerated approvals
7. Registered protocol: SPIMM-301, version 4.0, June 15, 2018
8. Registered statistical analysis plan: version 3.0, November 29, 2019
9. FDA prescribing information: Forzinity, revised September 2025

Review and corrections

Sources checked September 2, 2026 in Edmonton (September 3, 2026 UTC). Serum Research is responsible for this editorial summary and its corrections. No independent scientific, clinical, statistical, analytical, or regulatory review of this page is claimed. The published paper's fatigue-statistic discrepancy remains unresolved; this summary is not a reanalysis of participant-level data. It is educational, not medical advice or a preparation, administration, or dosing guide, and it makes no material-quality or equivalence determination.

Read how we review evidence, browse the study records or Research Library, and contact Serum Research about corrections.