Thymosin Alpha-1 TESTS Phase 3 Sepsis Trial: NCT02867267

The TESTS trial, NCT02867267, randomized 1,106 adults with sepsis at 22 centers in China to thymosin alpha-1 or matched placebo alongside standard care. In the corrected analysis, 28-day mortality was 23.4% with thymosin alpha-1 and 24.1% with placebo; the hazard ratio was 0.97, with a 95% confidence interval from 0.76 to 1.24. The trial did not show a reduction in its primary mortality outcome. A 2025 correction changed the survival estimates and must be read with the paper. This study does not establish benefit for a subgroup, regulatory approval, or equivalence to a separately sourced material.[1][2][4]

Study at a glance

The paper reports computer-generated blocks of eight, stratification by center and age, and an interactive allocation system. Investigators, participants, care providers, and statisticians were masked; emergency-unblinding arrangements existed. The registry labels the design quadruple-masked. These descriptions are compatible role labels, not proof that every possible unblinding was excluded.[1][3]

Endpoint hierarchy and analysis population

The paper identifies 28-day all-cause mortality as the prespecified primary endpoint. Secondary outcomes included 90-day mortality, ICU mortality, new infection, microorganism clearance, organ-support-free days, ICU and hospital stay, readmission, SOFA score, and quality of life. It classifies immune-marker changes as exploratory. The current registry instead lists mHLA-DR and regulatory T-cell measures among secondary outcomes; the two hierarchies are not identical.[1][3]

The study randomized 552 participants to thymosin alpha-1 and 554 to placebo. Seventeen withdrew consent before receiving any study drug, 10 and seven, respectively, so the modified intention-to-treat set contained 542 and 547 participants. That analyzed population is not the full randomized population.[1]

The paper reports center- and age-adjusted analyses, a Cox model for time to death, and a separate log-rank comparison. Except for subgroup-interaction testing, its confidence intervals and P values were not adjusted for multiple comparisons. Secondary and exploratory results therefore need nominal, nonconfirmatory interpretation. The paper attributes its methods to a prespecified analysis plan; the complete standalone protocol and statistical analysis plan were not obtained for this review, so their full content and amendment timing have not been independently verified.[1]

Corrected primary mortality result

Within 28 days, 127 of 542 participants in the thymosin alpha-1 group died, compared with 132 of 547 in the placebo group. That corresponds to 23.4% versus 24.1%.[1][4]

Two complementary estimates were reported:

Both confidence intervals include no difference. The hazard-ratio interval is compatible with a potentially meaningful reduction or increase in the hazard of death. The risk-difference P value and log-rank P value are different tests and should not be interchanged. The appropriate conclusion is that this trial did not demonstrate a mortality reduction, not that the interventions are proven equivalent.[1] The guides to confidence intervals and negative or null results explain those distinctions.

Why the 2025 correction matters

The BMJ issued a correction on May 30, 2025. It reports that pandemic-related disruption affected data verification at some centers, leaving outdated data in the originally published analysis. Survival information later obtained for nine of 1,106 randomized participants changed the 28-day hazard ratio from 0.99 (95% CI 0.77 to 1.27) to 0.97 (0.76 to 1.24). The 90-day hazard ratio changed from 0.94 (0.76 to 1.16) to 0.95 (0.77 to 1.17).[4]

The correction also reports lymphocyte-count errors for 35 participants and revisions to the data-sharing statement. The current full article retrieved through the official PMC service incorporates the corrected numbers. The primary conclusion remains null, but this data-verification history is material; an indexed correction is more informative than a generic “no retraction” statement.[1][2][4]

Secondary and exploratory outcomes

Ninety-day mortality was 31.0% with thymosin alpha-1 and 32.4% with placebo, with a corrected hazard ratio of 0.95, 95% CI 0.77 to 1.17, and log-rank P=0.61. The paper reports no statistically significant group difference for its other secondary clinical outcomes, including ICU mortality, new infection, pathogen clearance, SOFA score, readmission, lengths of stay, or organ-support-free days.[1]

Two exploratory change measures for the neutrophil-to-lymphocyte ratio differed at the nominal level: absolute change differed by −1.7, 95% CI −3.0 to −0.3, P=0.02; percentage change differed by −9 percentage points, 95% CI −15 to −2, P=0.006. These are related measures of one marker, not two independent demonstrations of benefit. The nominal day-7 assessment used end-of-treatment measurements when treatment ended earlier; it was not a uniform seven-day observation for everyone. Longitudinal immune-marker models did not establish a corresponding between-treatment pattern. These unadjusted biomarker comparisons cannot replace the null mortality endpoint.[1]

Subgroup findings are not confirmation

The paper reports treatment-by-subgroup interaction P values of 0.01 for age and 0.04 for diabetes. Participants younger than 60 had a hazard ratio of 1.67, 95% CI 1.04 to 2.67; those 60 or older had a hazard ratio of 0.81, 0.61 to 1.09. The diabetes subgroup estimate was 0.58, 0.35 to 0.99, while the no-diabetes estimate was 1.16, 0.87 to 1.53.[1]

However, the paper reports a Bonferroni threshold of P=0.006 for eight subgroup-interaction tests. Neither P=0.01 nor P=0.04 meets that threshold. The subgroup patterns, including the possible younger-adult harm signal, are hypothesis-generating; they do not establish a reliable subgroup benefit or harm estimate.[1] See subgroup analysis and multiplicity for why a within-subgroup P value does not establish a treatment-by-subgroup difference.

Adverse events

Within 90 days, at least one adverse event occurred in 360 of 542 participants (66.4%) with thymosin alpha-1 and 370 of 547 (67.6%) with placebo. The absolute difference was −1.2 percentage points, 95% CI −6.8 to +4.4, P=0.70. Serious adverse events occurred in 145 of 542 (26.8%) and 160 of 547 (29.3%), respectively, a difference of −2.5 points, 95% CI −7.8 to +2.8, P=0.38. These comparisons were not multiplicity-adjusted.[1]

Common events included anemia, fever, abdominal distension, and coagulation disorders. Investigators reported no unexpected serious adverse event related to thymosin alpha-1; that is an attribution, not a claim that serious events did not occur. Nonsignificant event-rate differences do not validate complete safety or exclude uncommon, delayed, population-specific, or differently exposed risks.[1] The adverse-event reporting guide explains the distinction between an observed event and assessed relatedness.

Attrition and missing data

The modified intention-to-treat analysis excluded 17 randomized participants who never received study drug after withdrawing consent. The paper describes this rule as prespecified, but their exclusion still matters when comparing the analyzed set with all 1,106 randomized people. After survival information was recovered for nine participants, the corrected paper states that the primary outcome had no missing data in the analyzed set; that does not provide outcome follow-up for every excluded randomized participant.[1][4]

Secondary outcomes had more missingness. The paper reports usable SF-36 quality-of-life records for 552 of 745 participants, or 74.1%; this is its stated assessment denominator, not a denominator reconstructed here from mortality counts. Missing secondary data used regression multiple imputation under a monotonic-pattern assumption. Imputation depends on assumptions and does not turn unavailable outcome records into direct observations.[1] See missing data and attrition for the resulting uncertainty.

The corrected data-sharing statement points to a Dryad dataset and says supplementary files contain analysis code. The repository's public API did not provide access to the dataset during this review. Participant-level data and code were not inspected or reanalyzed, and the complete separate protocol/analysis-plan file was not obtained. These are limits of this review, not proof that the materials do not exist.[1][4]

Sponsor, funding, and registry comparison

Sun Yat-sen University is the registered sponsor, and SciClone Pharmaceuticals is listed as a collaborator. The paper reports funding from the Sun Yat-sen University Clinical Research Program 5010, the Guangdong Clinical Research Center for Critical Care Medicine, and SciClone. It says the funders were not involved in design, collection, interpretation, or manuscript preparation. Several authors reported SciClone grants, and some reported consultancy fees involving SciClone or other companies.[1][3]

ClinicalTrials.gov lists the trial as completed with 1,106 participants, but has no posted summary-results module or attached protocol/analysis-plan documents. Its current record gives primary completion on January 22, 2021 and study completion on March 23, 2021; the paper describes conduct from September 2016 through December 2020. The registry was last updated April 10, 2023, before the publication and correction. These records should remain distinct; the registry cannot independently reproduce the corrected result table.[1][3][4]

How to interpret significance, effect size, and uncertainty

• Statistical significance: the primary mortality comparison was not statistically significant.
• Effect size: the observed absolute difference was less than one percentage point, much smaller than the eight-point difference used for sample-size planning.
• Uncertainty: the absolute-risk confidence interval does not include the eight-percentage-point mortality reduction used in planning, although it remains compatible with smaller benefit or harm.
• Clinical importance: mortality was the clinically important endpoint, and an exploratory immune-marker shift cannot replace it.

Important limitations

• The trial used a modified intention-to-treat set that excluded 17 randomized participants who received no study drug.
• The study was powered around an assumed eight-percentage-point mortality difference; it may not rule out smaller effects.
• Sepsis is heterogeneous, and the investigators could not determine exact time from sepsis onset to diagnosis.
• All 22 centers were in China, which can limit transfer to other care systems and populations.
• Secondary outcomes had variable missingness and relied on imputation.
• Secondary and exploratory comparisons were not adjusted for multiple testing.
• Subgroup interactions did not meet the study's Bonferroni threshold.
• A material post-publication correction changed survival estimates and immune-marker data.
• The registry has no posted results and cannot serve as an independent result replication.
• This review did not inspect the complete separate protocol/SAP or participant-level data and code.

What this study does not prove

NCT02867267 does not prove that thymosin alpha-1 lowers mortality in adults with sepsis. It does not establish benefit in an age or diabetes subgroup, because those interaction findings did not meet the multiplicity threshold and require confirmation. It does not establish efficacy for other infections, immune states, populations, durations, or outcomes.

The paper is also not an FDA approval record and does not authenticate any separately sourced material. A shared name, research-use label, certificate, or catalog listing cannot establish identity with the trial's controlled test article or reproduce its clinical exposure. The research material versus approved drug guide explains why these are separate questions.

Regulatory and material-identity boundaries

FDA's November 15, 2024 evaluation for the December 2024 compounding advisory meeting distinguishes thymosin alpha-1 free base from its acetate as different bulk drug substances. It states that neither was a component of an FDA-approved drug at that evaluation date. This advisory briefing is not a final regulatory determination and predates the 2025 TESTS publication and correction.[5]

On the source-check date below, FDA's current safety-risk page lists thymosin-alpha 1 among substances whose compounding nominations were withdrawn. It describes potential immunogenicity, peptide-impurity and characterization concerns, and inadequate information to determine harm under proposed compounding conditions. That listing is not a current Category 2 placement, a drug approval, or a finding that TESTS established safety for compounded or catalog materials.[6] Use the trial and FDA status-check guide to keep these records separate.

Frequently asked questions

What was the primary endpoint in TESTS?

It was all-cause mortality within 28 days after randomization.

Did thymosin alpha-1 reduce 28-day mortality?

No statistically significant reduction was demonstrated. Mortality was 23.4% versus 24.1%, with a corrected hazard ratio of 0.97 and a 95% confidence interval from 0.76 to 1.24.

Why are some older summaries different?

The BMJ corrected the paper in May 2025 after survival data for nine participants and lymphocyte-count errors were resolved. Current summaries should use the corrected estimates.

Did secondary outcomes show a clinical benefit?

The paper reports no statistically significant difference in its secondary clinical outcomes. The nominal neutrophil-to-lymphocyte-ratio change comparisons were exploratory and unadjusted for multiplicity; they do not demonstrate a clinical benefit.

Did the subgroup results prove benefit in older adults or people with diabetes?

No. The interaction P values did not meet the study's Bonferroni-adjusted threshold and remain hypothesis-generating.

Were adverse-event rates different?

The reported overall and serious adverse-event proportions were not statistically significantly different, but the trial cannot exclude every uncommon or delayed risk.

Does TESTS validate a commercial thymosin alpha-1 material?

No. The trial's result is specific to its controlled test article and protocol and does not establish the identity or equivalence of another material.

Primary and official sources

1. Primary corrected article: TESTS phase 3 trial; official PMC full-text record
2. PubMed: PMID 39814420
3. ClinicalTrials.gov: NCT02867267
4. BMJ correction: updated survival and immunological data; official PMC correction record
5. FDA: thymosin alpha-1-related bulk-substance evaluation, November 15, 2024
6. FDA: bulk substances that may present significant safety risks

Review and corrections

Sources checked September 2, 2026 (America/Edmonton) / September 3, 2026 (UTC). Serum Research is responsible for this editorial summary. No independent clinical, statistical, scientific, analytical, or regulatory review is claimed. This is not medical advice, a treatment recommendation, or a material-quality or equivalence conclusion.

Read how we review evidence, browse the study records or the wider research library, and contact us with a correction and its source.