Skip to main content
HYDROGEN STUDIESPOWERED BY ECHO WATER

Methodology

How studies get into this database, what we exclude, what a study page contains, how we label the strength of evidence, how summaries are drafted and checked, and what the database cannot tell you.

Scope

The database covers published research on molecular hydrogen (H₂) given to living things as an intervention: hydrogen-rich water, inhaled hydrogen gas, hydrogen-rich saline, hydrogen baths, and hydrogen-producing tablets or powders. It includes human studies, animal studies, cell and tissue studies, and reviews. As of November 2026 it holds about 2,300 papers published between 1948 and 2026. It does not cover hydrogen as a fuel or energy carrier, industrial hydrogen, or general oxidative-stress research in which hydrogen was not the intervention.

How studies enter the database

An automated search of PubMed, Europe PMC and CrossRef runs every six hours using a fixed set of hydrogen-therapy search terms. Each result is screened for topic before anything is stored: a filter removes hydrogen-energy and industrial papers, and records are matched on DOI and PubMed ID so a paper appears once. For every paper we store the original title, authors, journal, publication date, DOI and PubMed ID exactly as the source index reports them, and the study page links to that source. Study URLs are permanent once published.

What we exclude, and why

A paper is removed from the live database when it is retracted or carries a publisher's expression of concern, or when review shows it is off topic. Each removal records its reason. As of November 2026, 82 papers have been removed: 38 for retraction or publication notices and 44 for being off topic (general redox research without a hydrogen intervention, hydrogen-energy papers, and industrial fermentation studies). Removed pages return a "410 Gone" status so they leave search results rather than redirecting to another page.

What a study page contains

  • The original citation: title, authors, journal, year, DOI and PubMed ID, with a link to the paper.
  • A plain-language headline and summary written for a general reader, a short "key finding", and where it applies a practical takeaway.
  • An abstract excerpt of at most 300 characters followed by the link to the full abstract on PubMed or at the publisher. We do not republish full abstracts.
  • The evidence type (see below), with a plain warning when the study was done in animals or in a laboratory rather than in people.
  • Structured details where the paper reports them and we have extracted them: study design, sample size, duration, hydrogen concentration and delivery method, funding source and declared conflicts of interest. Coverage of these fields is partial and growing; as of November 2026 sample size is recorded for most human studies but funding and conflict-of-interest fields are recorded for a minority of papers. A field that was not reported is left blank, never guessed.
  • Links to the condition and body-system hubs the study belongs to, and related studies.

How we label the strength of evidence

Every study is classified by design: systematic review or meta-analysis, randomized controlled trial, controlled trial (non-randomized), uncontrolled trial (single-arm or pilot), observational study, case report, review or commentary, animal study, or laboratory study (cells or tissue). The label is shown on the study page. Animal and laboratory studies carry a visible note that results in animals or cells often do not carry over to people.

Only about one paper in seven is a human study (351 of about 2,300 as of November 2026), and the human studies are mostly small and short: the median trial has 23 participants and lasts about four weeks. Hub introductions therefore grade the overall evidence for a topic in GRADE-style terms, usually "Low" or "Very low", and say why: small samples, short durations, few replications, differences between trials, and risk of bias. A positive result in a single small trial is reported as exactly that.

How summaries are drafted and checked

Summaries, headlines and structured fields are drafted by large language models (currently Anthropic's Claude models, with OpenAI as a fallback) from the abstract or full text, then reviewed by a person before publication. Our reviewer, Sarah Taylor-Laine (M.Sc. Biology; Research Director at Echo Water), checks each study's summary and extracted details against the paper before it is published, whether she imports the paper herself or approves it from the queue of papers our automated finder discovers. Database-wide audits add a second check: in November 2026 all 2,305 headlines were audited and 674 rewritten, and extracted field values were spot-checked against the papers before being applied. Individual pages are corrected whenever a reader or researcher reports a problem. The editorial policy describes the review process and how to request a correction.

How hubs and guides are written

A hub introduction is written by the editorial team from the primary papers on that topic. It states what the human evidence shows, what animal data suggests, the doses and durations that were studied, and the open questions, with numbered citations that link to PubMed and to our study pages. Each hub shows an evidence grade and an "Updated" date. FAQ answers on a hub are checked against the cited studies. Guides are planned one at a time and reviewed before publication; nothing is generated in bulk.

Funding and conflicts of interest

Hydrogen Studies is funded by Echo Technologies LLC, a maker of hydrogen water products. Where a paper reports its funding source or the authors' conflicts of interest, we record them as structured fields and show industry funding as such, including any study funded by Echo Technologies. This extraction is in progress; as of November 2026 these two fields are populated for a minority of papers, so a blank field means "not yet recorded", not "no conflict".

What the database cannot tell you

  • The literature leans positive. Of the papers with a recorded outcome direction, about 94% report a positive result and about 6% report a neutral, mixed or negative one (November 2026). That reflects what gets published, not a verdict on hydrogen; small positive studies are more likely to be written up than null ones.
  • It is geographically concentrated. About 48% of the papers come from China and about 23% from Japan; the United States accounts for about 5%.
  • Human trials are small and short, as described above, and few have been replicated independently.
  • Summaries are machine-drafted and can be wrong in ways an audit has not yet caught. The linked paper is the source of record.
  • Citation counts come from OpenAlex, are shown with their refresh date, and are not a measure of quality.

The current figures behind each of these points are computed live on the limits of the evidence; the studies that found nothing are listed under null and negative results.

Changes to this methodology

This page is updated when the process changes; its last-updated date is shown at the top. Questions about the method, or requests for the underlying data, go through the contact page.