The first open guide to make each client a learning loop: clinics measure, intervene and re-measure; the records train biomedical world models; better models return sharper whole-life care
BEIJING, September 20, 2026 — DeepoMe, a Beijing-based startup, and the Beijing Tsinghua Industrial R&D Institute today launched the Good Healthspan Practice (GHP) Initiative on two channels at once: a concept paper, "When One Person Is the Trial: Toward Good Healthspan Practice (GHP), a Draft Good-Practice Guide for N-of-1 Evidence in Longevity Medicine" (Preprints.org, DOI: 10.20944/preprints202609.1682.v1), and ghp-ai.com, the guide's public home.
To the authors' knowledge, GHP is the first practice guide organized around a single question: what actually works for this one client? Researchers call the discipline N-of-1, or single-patient, evidence. In plain terms, the client's own before-and-after data serves as the control. The guide regulates no disease, no product and no laboratory. Its object is the quality of the reasoning behind every recommendation a longevity service makes.
The draft is not a finished code. It is a living baseline: free to adopt module by module, open for public comment through December 31, 2026, and built to change as adoption data accumulates. Each client loop leaves a written record: what was predicted, and what actually happened. Kept in one shared format across clinics, these records become the training data that biomedical world models now lack, and sharper models return better guidance to practice.
Longevity clinics are opening worldwide, yet no public document defines what a competent service looks like. Clinical guidelines govern the treatment of disease, not the conduct of a service, and wellness-industry certificates rarely reach scientific substance. The evidence problem runs deeper. A typical client carries several conditions at once and runs five to ten interventions in parallel, so population averages from conventional trials transfer poorly to that one person. The field therefore faces two gaps at once, in practice and in evidence, and the busier it becomes the wider they open. Reasoning for the individual is unavoidable in this work; left unregulated, it becomes arbitrary inference dressed as care.
GHP answers both gaps in one open, product-neutral design: six guides and thirty-two checkable criteria. Five process guides (GHP-100 Inquiry, GHP-200 Multi-omics Measurement, GHP-300 Root-Cause Attribution, GHP-400 Intervention and GHP-500 Contribution Review) define the full cycle of whole-life care. One requirement is hard: causal chains must reach a root-cause layer, starting with the hallmarks of aging, the small set of mechanisms the entire field shares. A sixth guide, GHP-600, runs across all five and makes individualized causal inference itself checkable. Every causal claim must be registered, falsifiable and honestly reviewed. Adoption is free, voluntary and by self-declaration, module by module, with no external audits and no certificates.
The seven criteria of GHP-600 borrow tools that regulators have already validated: single-patient (N-of-1) arrangements used by the FDA, real-world evidence frameworks, and the methods of learning health systems. The guide turns them into clauses a clinic can be checked against. All of them rest on one question, the counterfactual: what would this marker have done without the intervention? A large clinical trial answers by averaging thousands of people. An N-of-1 design answers by lending the person their own untreated periods as the comparison. GHP, in effect, formalizes what rigorous practitioners already do informally: measure, intervene, re-measure. It adds the discipline that makes the result count as evidence.
The timing is not accidental. Biomedical world models, AI systems built to simulate and steer what happens inside one organism, were established this September as a serious scientific direction in a Cell Perspective (DOI: 10.1016/j.cell.2026.08.032). DeepoMe's SteeraMed framework and its N-of-1 intervention-reasoning preprints are earlier work on the same line. Models of this kind lack one input above all: records of what happened to a real person under a real intervention, predicted in advance and checked after the fact. That is exactly what every GHP-compliant client loop produces. As the models improve at predicting what an intervention will do, and at working out how much each intervention contributed, each cycle leaves less uncertainty for the next client.
"Longevity medicine generates more causal claims per patient than any field in medicine, with the least machinery for checking them," said Dr. Jianghui Xiong, founder and chief scientist of DeepoMe. "GHP regulates the reasoning itself — not the disease, not the product, but the inference behind every recommendation."
"The industry does not need another concept white paper; it needs clauses a clinic can be checked against," said Junxiang Zhang of the Industrial Ecology Research Center, Beijing Tsinghua Industrial R&D Institute. "We open-sourced the text so the whole field can take it apart."
The guide also defines interfaces for three classes of ecosystem partner. Assay vendors supply the molecular anchors, data platforms carry the records, and AI developers get structured data to train on. A deliberately short list of shared mechanisms keeps these interfaces open without diluting focus. Records kept in one shared format are more than good practice; they are reusable data assets.
GHP is not a standard under any national standardization law, not a certification and not an association. It is a public working document (v0.2), open for community comment through December 31, 2026 at ghp-ai.com, where an open registry lists adopted modules and self-authored client-loop case stories. The paper is available at https://www.preprints.org/manuscript/202609.1682 (DOI: 10.20944/preprints202609.1682.v1).
About GHP
Good Healthspan Practice (GHP) is an open, product-neutral working draft of a good-practice guide for longevity-medicine and health-management services. Six guides cover the full service cycle: inquiry, multi-omics measurement, root-cause attribution, intervention and contribution review, plus GHP-600, which makes individualized causal inference registered, falsifiable and open to audit. Adoption is free, voluntary and by self-declaration. The draft text (v0.2), the open registry and the public comment window (through December 31, 2026) are available at ghp-ai.com.
About DeepoMe
DeepoMe Limited is a Beijing-based startup building molecular-level aging measurement and AI-driven root-cause analysis for longevity medicine, including the Capome multi-dimensional aging assay, the SEMO precision-nutrition algorithm and the SteeraMed steerable world-model framework (https://steeramed.com). https://deepome.com
About the Beijing Tsinghua Industrial R&D Institute
The Beijing Tsinghua Industrial R&D Institute is an innovation platform affiliated with Tsinghua University bridging scientific research and industrial application.
Media Contact
DeepoMe Media Relations
Email: info@deepome.com
Websites: https://deepome.com | https://ghp-ai.com
