Methodology

This page explains openly how this database is built — from selecting the studies to checking their reliability. The goal behind it: you should be able to verify every claim yourself.

Human studies only

The focus is exclusively on human studies of molecular hydrogen (H₂) — research in which actual people took part. Animal and cell experiments are scientifically important, but they don't yet tell us whether something also works in humans, so they are kept separate.

Many papers contain the word „hydrogen" but mean something entirely different — such as hydrogen sulfide (H₂S), hydrogen peroxide, the „hydrogen" in medical breath tests, or hydrogen as a fuel. We filter those out so that only molecular hydrogen (H₂) appears here.

Where the studies come from

The basis is PubMed — the official, free literature database run by the U.S. National Library of Medicine (part of the NIH health agency). It indexes millions of peer-reviewed papers. Each paper carries a unique number, its PMID. We pull the H₂ human studies from there automatically, remove duplicates, and link every study back to its PubMed record — so you reach the original source with one click.

We keep the collection current and, alongside PubMed, also search ClinicalTrials.gov, Europe PMC and Google Scholar. Is a study missing? Submit it here — we review every tip.

Delivery methods

Each study is classified by how the hydrogen was given: drinking hydrogen-rich water (HRW), inhalation of H₂ gas, tablets (effervescent tablets that release H₂ into water), saline / infusion (IV), or bath / topical through the skin. Click a delivery method to see all of its studies:

How strong is the evidence?

Not every study carries the same weight. We tag each one with an evidence level so you can see at a glance how solid a result is:

• RCT — the „gold standard": participants are split at random into an H₂ group and a comparison group (often with a placebo). This makes it most likely that any difference is due to the hydrogen — and not to chance or expectation.
• Pilot / Observational — smaller, early, or observational studies, often without a comparison group. Interesting signals, but weaker: other explanations can't be ruled out.
• Review / Meta-analysis — studies of studies: they pool many individual trials. They can give the best overall picture — but they are only as good as the studies they summarize.
• Mechanism / Preclinical — lab and animal work explaining how H₂ might act. Foundational, but not proof of a benefit in humans.

Note: PubMed does not tag every randomized trial as such, so the RCT count shown is a cautious lower bound. A finer GRADE-style quality assessment (sample size, risk of bias, statistical significance) is in preparation.

How we check reliability

So that you can trust the data — and check it yourself — we verify on several levels:

• Every entry links to the original source. Each study links to its PubMed record; you don't have to take our word for anything — you can read it.
• Every DOI is verified against CrossRef. A DOI („Digital Object Identifier") is like the permanent licence plate of a scientific paper — a code that always leads to exactly the same work, even if web addresses change. CrossRef is the official, non-profit registry that issues these DOIs for most publishers. We check every single DOI against CrossRef to confirm it really exists and points to the correct paper. That means: no fabricated and no dead references — every link here is genuine and verifiable.
• Abstracts shown verbatim. We show the scientific summary (the „abstract") in its original wording; the authoritative source always remains the PubMed link.
• Null and negative results shown just as prominently. This is our most important honesty rule: studies that found no effect — or even an unfavourable one — are listed and described just as clearly as positive ones. Sellers and fan sites usually show only the „it works!" studies and hide the rest — that's called cherry-picking. We deliberately do the opposite, so you see the complete, honest picture and can form your own judgement.
• Limitations stated honestly. On every study page we name the weaknesses — such as few participants, no control group, short duration, or unclear dosing — so you can judge how much a result is worth.
• No overstatement, no selling. This site sells nothing, runs no advertising and makes no health promises. You can read who is behind this work on the About page.

Verify it yourself with an AI assistant (open data, API & MCP)

You don't have to search this site by hand: you can ask your own AI assistant (such as ChatGPT, Claude or Perplexity) to search and cross-check the data here. The entire database is available openly and machine-readably — no sign-up, and deliberately token-efficient: an agent loads the whole catalog in a single request instead of clicking through hundreds of pages.

• Open JSON API. /api/meta.json (overview & filters), /api/catalog.json (all studies, compact) and /api/studies.json (full text per study) — each with PubMed link and DOI.
• MCP server for AI agents. At https://mcp.h2medicine.org an MCP server offers tools such as search_studies and get_study; MCP-capable assistants query the database directly. Technically inclined users can reach the same data from the command line (e.g. with curl).

For example, tell your chatbot “Use h2medicine.org/api/catalog.json and list all human studies on diabetes with their source” — or connect an MCP-capable assistant to https://mcp.h2medicine.org. Every answer can then be checked against its PubMed/DOI link.