Health Record Rights Index

io.github.evil-robotv1.0.0更新于 Oct 4, 2026

Health record rights: 198 countries scored 0 to 100 with sources; the 50 states and DC; real cases.

已验证Streamable HTTP可网页运行Data & AnalyticsKnowledge & Memory

概览

AI 生成的概览

让助手查询一个只读指数，涵盖 198 个国家和美国各州在健康记录权利方面的评分，并附引用来源。

功能
提供对健康记录权利指数的只读访问：198 个国家和地区在八个加权类别下的 0 到 100 分评分，另有覆盖 50 个州和哥伦比亚特区的美国州级数据层。同一数据也通过该项目的 API 以开放 JSON 形式提供；README 还介绍了各国数据文件、导出文件、评分标准、论文，以及跨境记录交换的“旅行患者”视图。MCP 端点被描述为提供与 API 相同的数据。
适用场景
当助手需要回答有关健康数据权利、患者访问与同意，或国家与美国各州在记录控制方面如何比较的问题，并且需要引用公开来源时，适合使用。它是研究参考，不是法律建议。
运行要求
一个远程 MCP 端点；未声明需要软件包、密钥或注册。README 说明 API 与 MCP 服务器按地址共享每分钟 60 次请求的限制，突发可达 120 次，批量使用应改用下载文件。
安装前请注意
该数据是评分研究指数，不是法律建议；README 指出尚无人类评分者为其打分，且部分分数无法精确复现。它披露发布方销售健康数据验证产品，作者存在商业利益，包括一家健康信息交换机构和一家检测服务关联公司。数据采用 CC BY 4.0，需署名；引用的法律条文和第三方标题不在许可范围内。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Health Record Rights Index，将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Web executable，通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "health-record-rights-index": {
      "type": "http",
      "url": "https://healthrecordrights.com/mcp"
    }
  }
}

README

Health Record Rights Index

Who holds the record? An open index that scores 198 countries and territories, from 0 to 100, on one question: can a person see, control and share their own health record?

By SuperTruth Inc. Index version 1.0, data as of 2 October 2026. Data release 1.1 adds the US state layer: what the law of each of the 50 states and DC adds to federal law, read on 3 and 4 October 2026. Data and paper CC BY 4.0, code MIT.

Use the data

The Health Record Rights Index is also open as read-only JSON at https://healthrecordrights.com/api/v1/, with no key and no sign-up. The same data is open to AI assistants through an MCP server at https://healthrecordrights.com/mcp. MCP is a shared standard that lets an assistant read outside data. Each address may make 60 requests a minute, with bursts up to 120, shared between the API and the MCP server. For the whole index at once, use the downloads instead. The data is licensed under CC BY 4.0; credit "SuperTruth, Inc., Health Record Rights Index". Quoted law and source titles stay with their owners, and the code in this repository is MIT. Every route, field and example is on https://healthrecordrights.com/developers/.

[Health Record Rights Index: the globe and the headline finding]

The finding

In 2026, none of the 198 countries and territories we rated puts a person fully in charge of their own health record.

  • No country holds the top band. No country reaches Leading (85 and up). By band, 9 rate Strong, 66 Mixed, 111 Weak and 12 Poor. The median is 39.
  • The leaders are a group, not a winner. Finland, Denmark, Estonia, Hungary and Sweden have the highest scores. Allowing for scoring error, any of them could rank first, so we name them together.
  • Someone else holds the keys everywhere. By default the record sits with the state or with providers: 51 systems are Shared (the person has a working choice inside a shared system), 81 Institutional and 66 State. None is Individual.
  • Ranks are ranges. Every country carries the range of ranks it holds in 90% of simulations. Read those, not single positions.

Coverage: all 193 UN member states, plus Greenland, Kosovo, Palestine, Taiwan and Vatican City. Including a place takes no position on its status; we cover every place that runs its own health system.

Top 10

RankLikely rangeCountryScoreBand
11 to 6Finland71Strong
21 to 8Denmark70Strong
3=1 to 13Estonia68Strong
3=1 to 13Hungary68Strong
51 to 17Sweden67Strong
6=2 to 20Australia66Strong
6=2 to 20France66Strong
6=2 to 20Norway66Strong
93 to 23Austria65Strong
103 to 26Portugal64Mixed

Other large systems: Germany 63 (11=), the Netherlands 60 (23=), the United Kingdom 59 (27=), Japan 56 (38=), Canada and Brazil 52 (47=), the United States 48 (55=, likely 45 to 80), India 45 (71=), China 43 (80=). The full ranking is in exports/who-holds-the-record-scores.csv and Table 2 of the paper.

[Every country in rank order, with its score interval and rank range]

Every country in rank order. The lower panel draws each country's rank range; the upper panel adds the score intervals behind them. No country reaches Leading.

What we measure

Eight weighted categories, each scored 0 to 100 against a written rubric (RUBRIC.md):

CategoryWeight
Patient access to the full record20%
Patient control and consent20%
Privacy and security15%
Connected care journey15%
Protection from commercial use10%
Clinician access at the point of care10%
Research and trial consent5%
Clinical AI governance5%

The overall score is the weighted sum. The weights are the authors' judgment; the paper tests how much the ranking depends on them. Each country is also placed in one of four classes for who holds the keys by default:

[Who holds the keys: the four classes]

[Scores in each of the eight categories]

Which rights lag. Patient control and clinical AI governance sit lowest; research consent heaps just below the middle of the scale, which is why it barely separates countries.

How it was built

[How the index was built, in six stages]

  1. Research. One research agent per country, in the country's own languages, using public sources only: laws, government and regulator pages, court decisions and published news. The rubric requires every cited source to be opened. The agents are built on Anthropic's Claude (Claude Opus 5.5 through Claude Code, 30 September to 2 October 2026). The paper discloses this in full.
  2. Cross-check. A second agent session of the same model family checks each country file against the rubric anchors and against peer countries with the same legal situation.
  3. Review and audit. Agents audit consistency across countries; the authors rule on every finding and every score change, and every change is logged in docs/SCORE_CHANGES.md.
  4. Scoring. Weighted sum, then five bands.
  5. Testing. Ranking stability (10,000 simulations), two blind re-scorings, a full second reading of 43 countries, comparison with outside measures, and a link check of every cited source.
  6. Publication. Scores, every source, a one-page brief per country, and the paper.

Everything comes from public information: 5,147 cited sources. Of 7,156 cited links, 7,011 opened on recheck on 2 October 2026; 88 were blocked by bot protection and 57 did not answer. No link was dead.

The US state layer

The index scores the United States as one country, on federal law. The state layer records what each state's own law adds to that federal floor, for the 50 states and DC: copy rights, consent for health information exchange, health data held outside HIPAA, breach notice, a private right to sue, rules for AI in care, and four kinds of sensitive record (genetic, HIV, mental health and reproductive). It gives no state score and no state rank, and it changes no country score.

Each state file in data/us-states/ cites the provisions its answers rest on: 929 in all, each with its citation, its official link and the words of the law quoted from it. 912 of the quotes were found word for word in a stored copy of the page (quoteVerified: "raw", with the copy's SHA-256 in rawSha256). The other 17 come from pages built by script that could not be stored as text (quoteVerified: "session"); a second agent session re-read each of them. The stored copies are not in this repository, because statute pages hosted by vendors carry their own terms.

A blind re-reading of 150 sampled cells, by agent sessions that could read only stored copies of the cited pages, agreed with the published cells at an intraclass correlation of 0.965 on the 114 cells with ordered levels where the rater settled a level (bootstrap interval clustered by state, 0.92 to 1.00). Section 11 of the paper sets out the method and its limits. The plan, the rubric, the log of changes, the decision rules, the rulings and the agreement results are in analysis/us/states/.

This is not legal advice. Each cell records what a primary legal text said on the date it was read.

How far to trust it

We published the checks against ourselves, including the ones that do not flatter us.

  • Two blind re-scorings. Separate agent sessions re-scored random samples of cells from the cited sources alone, without seeing our scores or our site. On the first 65 countries, agreement was 0.82 (intraclass correlation). On a pre-registered sample of 136 cells from the 133 countries added on 2 October, agreement was 0.84. Access to the record agreed least (0.68), mostly where the rater could not open a cited law text, so a single access score for an added country is not reproducible to the point.
  • No human rater has scored the index yet. That is the next step.
  • The lead group is on a rule set before the results. A country is in it if it ranks first, alone or tied, in at least 5% of simulations. Sweden is on the edge: 5.09% under the published run, 13 of 16 random seeds, and 2.8% if ties are not counted.

[Share of draws ranked first, by random seed]

  • Confidence labels follow a printed rule. High means at least 7 of 8 categories rest on an official source with no fact marked unchecked; low means 4 or more do not. 27 countries are high, 105 medium and 66 low. The label is about the evidence, not the country's rights, and it never changes a score.
  • Income matters but does not decide. Richer countries score higher, and income explains about half of the spread, but not all of it.

[Overall score against GDP per head]

The traveling patient

Does your record follow you across a border? The live page at https://healthrecordrights.com/traveler/ lets you pick where you live and a trip, and shows which of the five parts of your record (health summary, medicines, lab results, scans and X-rays, hospital notes) reach a doctor in each country by themselves, and which you would have to carry as a copy.

  • Of 702 one-way routes between EU countries, a health summary travels by itself on 143.
  • No source says the EU's cross-border service also works for people who move, not just visit.
  • 138 of 198 countries give a person a right to a copy of their record.
  • EU law adds a right to ask for each part to be sent to another member state: the health summary and medicines from 26 March 2029, the rest from 26 March 2031. A right in law does not move a record today.

[The three worked trips, stop by stop]

[What EU law adds for each part of the record, and when]

What is in this repository

PathWhat it holds
data/One JSON file per country: the eight category scores with summaries and detail, the laws, who holds the keys, and every cited source with its date
data/us-states/One JSON file for each of the 50 states and DC: what the state's law adds to federal law, question by question, with every provision cited and quoted
exports/The country data as CSV and JSON for analysis
RUBRIC.mdThe scoring rubric, its anchors and the confidence rule
DECISION_RULES.mdThe rules the build enforces, each with its check (for example, no single country is called first; the leaders are named as a group)
docs/Score change log, stories rules, evidence grade method
paper/The working paper (PDF and Markdown), its stylesheet, build script and figures
analysis/robustness/Ranking stability, lead-group rule and the seed sweep
analysis/reliability/, analysis/reliability2/The two blind re-scoring studies: plans with their hashes, rater kits, samples, ratings, results and reports
analysis/external/Comparisons with the WHO Global Digital Health Monitor, Eurostat and income
analysis/crossborder/Cross-border record exchange and copy rights behind the traveling patient
analysis/confidence/, analysis/dti/, analysis/variance/, analysis/strain/Confidence labels, evidence grades, variance structure, health system strain
analysis/literature/Literature review
analysis/us/states/The US state layer: plan and its hashes (PLAN.md, plan_stamp.txt), rubric (RUBRIC_STATES.md), log of changes to the plan (DEVIATIONS.md), decision rules, counts by question (aggregate.json), how rulings were made and every ruling applied (main/RULINGS_METHOD.md, main/rulings_applied.json), and the agreement results before and after the rulings (main/reliability.json, main/reliability_post_rulings.json)
LICENSE-DATA.mdWhat the CC BY 4.0 data license covers, and what it does not (quoted law, third-party titles, linked pages)
scripts/Paper generation (every number in the paper is computed, never typed), figure generators, checks and tests
build.js, template.html, brief.js, presskit.js, nav.js, traveller*.jsThe code that builds the site from the data

Reproduce it

sh
npm installnode build.js                  # builds the site into out/node scripts/paper_gen.js      # recomputes every number in the paper; refuses stale inputszsh paper/build.sh             # builds paper/paper.pdf (needs pandoc and Chrome)

The paper generator checks a hash of the data against every analysis output and stops if any figure was computed on older data.

One limit: the published accounts (the stories layer) are not in this repository, so that removal requests can be honoured. build.js runs without them and leaves the stories out; paper_gen.js needs them for Section 5's counts and will not run from this repository alone. Every score, rank and analysis rebuilds from what is here. Two more limits for data release 1.1: the code here is the version 1.0 code, so it rebuilds the country data but not the state pages or the version 1.2 paper; and the state layer's harness scripts are not in this repository, so the hashes in analysis/us/states/plan_stamp.txt cannot be checked from here.

How to cite

Snyder, J. A., Hill, R. P., IV, Raney, D., & Sims, L. (2026). The Health Record Rights Index: Who Holds the Record in 198 Countries? (version 1.2). SuperTruth Inc. https://doi.org/10.5281/zenodo.23137550

Paper, version 1.2: https://doi.org/10.5281/zenodo.23137550 (all versions: https://doi.org/10.5281/zenodo.23120174) Data, version 1.1: https://doi.org/10.5281/zenodo.23137895 (all versions: https://doi.org/10.5281/zenodo.23120172)

Citation metadata is in CITATION.cff. The four authors share equal credit.

Licences

  • Data, scores and paper: CC BY 4.0. Credit SuperTruth and link the index. LICENSE-DATA.md says what that covers: quoted law, titles of others' pages and the linked pages keep their owners' terms.
  • Code: MIT (see LICENSE).
  • The SuperTruth name and logos in brand/ are trademarks of SuperTruth Inc. and are not covered by either licence.

Disclosure and corrections

SuperTruth sells health data verification products, and three of the authors are its officers. The fourth, Leann Sims, is Chief Executive Officer of MedSync, a US health information exchange that works with SuperTruth; the state layer asks about consent for health information exchange and does not name or score MedSync or any other exchange. imaware, an affiliate of SuperTruth, provides testing services under agreements with some US state governments; the index scores the United States as one country, and the state layer gives no state a score or a rank. We built this index and chose its stories ourselves. No government, health system, company or person paid to be included or had any say in the scores.

Press and interviews: Rheanna Crescenzo, [email protected]

This is a research tool, not legal advice. If we got a country wrong, open an issue with the page that shows it and we will fix it in the open.

The published accounts of record problems shown on the site are not deposited here, so that a removal request can be honoured; each links to its public source on the site.

来源：README.md，提交 fc82ff3

工具

0
工具元数据尚未被收录。

版本历史

1
  1. v1.0.0最新Oct 4, 2026