When One Person is the Trial: Toward Good Healthspan Practice (GHP), a Draft Good-Practice Guide for N-of-1 Evidence in Longevity Medicine

1. Background

Longevity medicine has moved from the laboratory to the storefront. Consumer interest in biological-age measurement and anti-aging regimens is high, institutions offering these services are proliferating, and capital and franchises are following. What the sector lacks is what every mature health-adjacent industry acquired before it scaled: a public, verifiable definition of competent practice. Without such a definition, a consumer cannot tell a rigorous program from a supplement bundle with a biological-age number attached; an institution that invests in rigor cannot signal that investment; and regulators and payers are left without any reference against which to evaluate claims.

Other parts of medicine solved this problem through the GxP family: Good Manufacturing Practice (GMP) for pharmaceutical production, Good Clinical Practice (GCP) for trials, Good Laboratory Practice (GLP) for non-clinical studies, and Good Pharmacovigilance Practice (GVP) for drug-safety surveillance [1]. Every one of them emerged when an activity became safety-critical and trust-critical at scale, and longevity medicine has now reached exactly that point without a GxP of its own.

A second problem sits underneath the first: the people this field serves have three properties that conventional evidence tools were never built for. The typical client is multi-morbid, and the field's goal is health span, meaning function maintained across conditions rather than the cure of any single disease [2,3]. Practice, in addition, involves many interventions at once, with nutrition, exercise, supplementation and medical technologies running in parallel five to ten at a time, so even the best randomized controlled trials (RCTs) rarely disentangle their individual contributions. Nutritional and metabolic complexity, finally, varies so deeply between persons that no single model transfers intact from one individual to the next (Figure 1).

Evidence-based medicine handles heterogeneity by classification: patients are sorted into homogeneous diagnostic boxes, and a guideline derived from population statistics is applied box by box [4]. Longevity medicine breaks this pipeline at its first step, because multi-morbidity makes the boxes overlap and parallel interventions break the clean attribution on which box-level evidence rests. What remains for the practitioner is inference, from mechanism, from measurement, from response. Regulators have arrived at the same conclusion from the other side: the FDA's arrangements for single-patient (N-of-1) trials are an explicit accommodation to precisely this situation [5]. A second development makes this the right moment: biomedical world models, AI systems built to simulate and steer what happens inside one organism, have arrived as a serious scientific direction [6]. The result is a sector where inference is unavoidable but unregulated. Every program implicitly makes causal claims, and nothing requires those claims to be registered, falsifiable, or honestly reviewed. This article therefore takes a first step in that direction, proposing a working draft of Good Healthspan Practice (GHP), an open good-practice guide organized in modules, with root-cause attribution at its center. Under this draft, individualized causal claims become registered, falsifiable and honestly reviewed. The draft is released for community discussion and iteration, not as a finished code for adoption.

2. The Dual Gap

We formalize the sector's problem as two gaps (Figure 2).

Practice gap (what to do). No public document defines what a competent longevity-medicine service looks like. Clinical guidelines govern the diagnosis and treatment of diseases, not the conduct of a service, and wellness-industry quality certificates rarely reach scientific substance. An institution therefore has no public reference to check its practice against, and a consumer cannot tell a rigorous program from a loose one. What is missing is concrete: agreed expectations for inquiry, measurement, interpretation, intervention and review.

Evidence gap (how to reason for one person). Textbook evidence does not answer the questions this field faces. The typical client is multi-morbid and takes several interventions at once. Large trials enroll homogeneous patients and usually test one intervention at a time, so their average results extrapolate poorly to such a person. The practice must still decide, so reasoning from mechanism and individual data is unavoidable. Left unregulated, it becomes arbitrary inference dressed as care. What is missing is a disciplined version of that reasoning: the individualized evidence generation this guide proposes below.

Standards that address only the first gap would produce well-documented arbitrariness, and the field cannot wait for more trials to close the second gap, because the clients are here today.

3. The GHP Design

GHP (Good Healthspan Practice) is an open, product-neutral good-practice guide issued as a voluntary initiative. In its current form it is a public working document that institutions can use and check themselves against, distinct from formal standards, certifications and associations. Its v0.2 text organizes criteria along two tracks, one for each gap (Figure 3; Box 1): five process guides for the service cycle answer the practice gap, and GHP-600, which runs across all five, answers the evidence gap.

Box 1. The six guides (five process + one that runs across all of them). 

Code	Guide	Core requirement (summary)	Criteria
GHP-100	Inquiry	Four-layer causal history (exposure → root cause → function → phenotype); structured family history; East–West integrated history-taking	6
GHP-200	Multi-omics measurement	Integrated multi-omics with ≥1 molecular-level aging anchor (e.g., a DNA methylation clock); harmonized, lag-aware re-testing	4
GHP-300	Root-cause attribution	Hard requirement: causal chains must reach the root-cause layer (first candidate: hallmarks of aging); stopping at functional imbalance does not qualify. Inference mandatory, method-free; hallmark measurement a bonus, never mandated	5
GHP-400	Intervention	Root-first graded prescriptions; one theme per cycle; expectations and lag windows pre-registered; executable to dose and frequency	5
GHP-500	Contribution review	Same-assay re-test alignment with significance gating; per-intervention contribution ranking with confidence grades; honest reporting of unexplained change	5
GHP-600	Inference integrity (runs across all five)	Individualized causal inference that is registered, testable and open to audit	7

Scope is bounded by design. GHP applies to the service practice of longevity-medicine and health-management institutions: the full cycle of inquiry, measurement, interpretation, intervention and review, or any single module thereof. Diagnosis-and-treatment guidelines belong to medical societies; pharmaceutical production and clinical trials belong to GMP, GCP, GLP and GVP. GHP stays within service practice and states this boundary in its front matter.

Four design decisions deserve emphasis.

Root-cause positioning as a hard requirement (GHP-300). Any interpretation of test data is, at bottom, an inference, and interpretations that stop at "functional imbalance" leave intervention surface-level, so GHP requires the causal chain to be positioned at the root-cause layer, with the hallmarks of aging as the first candidate set [7,8]. The logic is simple: a mechanism studied in many people by the whole field carries high consensus, while a mechanism seen only in one clinic visit carries little. The hallmarks are the first kind — a small set of shared mechanisms, built by decades of population research — so reasoning that runs through them is anchored to what the field already knows. The guide treats them as working candidates rather than settled causes, and individual predictions still decide. Interventions built around a few shared mechanisms are also easier to pre-register, monitor and review. The method is deliberately left open, and molecular measurement is treated as a bonus rather than a mandate. The guide must not become a sales channel for any assay.

One theme per cycle, with interventions in parallel (GHP-400). Real programs run five to ten interventions at once, and GHP does not forbid that. What it asks is that each cycle be organized around one dominant theme, that expectations for each intervention be pre-registered, and that at least one core intervention carry within-person control (criterion 600.4). Attribution then has a fighting chance even under parallel practice.

No lock-in to any test supplier (GHP-200). The criterion is at least one molecular-level biological-age measurement, for example a DNA methylation clock; any qualified third-party laboratory satisfies it. Re-testing windows follow biological lag rather than commercial convenience.

Process criteria are self-check documents with two reference tiers. Each of the five process guides carries an adoption tier and an advanced tier, both intended for institutional self-assessment rather than external grading. For example, GHP-400 adoption calls for goal binding, pre-registration and executability (criteria 400.2, 400.4 and 400.5); the advanced tier calls for all five. Adoption works by self-declaration: an institution checks its own practice against the guide, states publicly which modules it has adopted, and the secretariat records that statement in an open registry. Institutions may voluntarily share one complete client closed loop (inquiry → measurement → interpretation → intervention → re-test review) as a self-authored case story; the registry lists these stories so that others can read how the guide works in practice.

4. GHP-600: Making One-Person Inference Checkable

GHP-600 is, to our knowledge, the first practice guide whose object is the quality of individualized causal inference itself (Figure 4; Box 2). Its seven criteria translate the tools regulators have already developed for N-of-1 trials, real-world evidence and learning health systems [9,10,11] into routine longevity practice. What those tools share is a way of managing uncertainty: when a claim cannot wait for a population trial, the discipline moves to predicting, documenting and checking what happens in this one person. GHP-600 carries that same discipline into daily clinic work. Beneath all the criteria sits one question, the counterfactual: what would this marker have done without the intervention? A group trial answers it by averaging across randomized people; an N-of-1 design answers it by lending the person their own untreated periods as the comparison. Managing that uncertainty, honestly and in the open, is the bedrock of the whole guide. Like the five process guides, it applies to health-management services, to the records a service keeps and the reasoning behind them; it does not enter the exam room, where diagnosis and treatment remain with clinical medicine.

Box 2. GHP-600 criteria. 

#	Criterion
1	Hypothesis registry. At least three root-cause-layer hypotheses (hallmarks direction), each with two or more classes of prior sources (family and genetics, lifestyle, environment, prior testing, current testing, dynamic monitoring) and a counted confidence grade: A requires four or more independent sources, B exactly three, C two or fewer. Unresolved contrary evidence forces demotion or exclusion. Grading counts sources, not rhetorical confidence; the counting rules are public, while the exact thresholds will be fixed by platform data in later volumes.
2	Falsifiable predictions. Each hypothesis carries at least one prediction of the form "indicator X, within lag window W, direction D". Failed predictions demote the hypothesis. Post-hoc rationalization is prohibited.
3	Pre-registered lag windows. Expected effect windows are registered at issuance (for example, strength and cardiorespiratory outcomes 8–12 weeks; metabolic and weight outcomes 12–24 weeks; methylation clocks 12 weeks or longer; illustrative values, to be formalized in the next volume). Out-of-window re-tests are excluded from evaluation.
4	N-of-1 structure. At least one core intervention runs as a self-comparison for that client: measure at baseline, treat, measure again, and where safe, pause once to measure what returns. The client's own untreated state serves as the control, which is exactly what group trials cannot give. When pausing is unsafe, as with an essential nutrient, the dose is stepped or the intervention alternates on and off in a fixed rhythm. Analysis follows the N-of-1 methodological literature [9,12]: the person's on-treatment and off-treatment periods are compared within the same individual.
5	Inference transparency. Disclosure covers the input evidence set, the inference path, uncertainty quantification, and a three-tier evidence label: tier I population evidence (RCT or meta-analysis), tier II mechanistic prior plus individual evidence, tier III expert opinion. Tier III alone cannot support a root-cause hypothesis. Mixed chains declare the lowest tier. Tiers describe the kind of evidence, not which kind answers an individual question. Where population evidence is absent, claims must say "inferred" and never impersonate "evidence-based".
6	Multi-morbidity. With two or more coexisting conditions, a shared-root-cause analysis (root-to-phenotype mapping with coverage) is mandatory. Per-disease plan stacking is prohibited.
7	Honest review. Contribution decomposition carries confidence grades. The fraction of unexplained change must be reported. Shared improvement is not credited to any single intervention. Findings roll into the next cycle's hypothesis registry.

The stance is deliberately two-sided. Against the view that only large trials count, GHP-600 asserts that heterogeneous, multi-morbid individuals cannot wait for extrapolable RCTs [13,14], and that mechanistic reasoning combined with individual data is legitimate evidence when labelled as such. Against unregulated guesswork, it asserts that inference without registration, falsifiability and honest review is not medicine but marketing. The one-sentence ethic: allow inference; forbid packaging.

Read deductively, the argument runs as follows. Because the object of care is multi-morbid and receives many interventions at once, evidence must be generated within the person. Because within-person evidence is inference, inference must be disciplined. Because disciplined inference must be verifiable by a third party, it must be registered, falsifiable and open to audit. GHP-600 is therefore not an arbitrary design choice; it is the shape that standardization tends to take once the object of care is what it is.

5. Governance and Neutrality

GHP is published as an open initiative: a public technical document, multi-party endorsement, and self-declaration-based adoption. Anyone can read the guide, comment on it and declare adoption; participation is free. Founding endorsement is structured across four sectors: one research institution; two to three clinical experts in personal capacity; one or two industry participants at deliberately low weight with disclosure; and three to five practice institutions. A secretariat holds text editorship on the W3C model (the World Wide Web Consortium, which maintains the web's public standards). A staged evolution path leads from the unincorporated initiative toward formal standards published by registered societies (in China, for example, the T/-numbered team standards) as adoption grows.

Neutrality is built into the structure. Guide texts describe what a practice should do; any product or laboratory that meets the requirement qualifies, and an implementations page lists tool paths and generic, product-free paths side by side. Adopting the guide costs nothing. The secretariat's own commercial interests are disclosed here and in the guide's front matter. All adopters appear in the registry on the same footing: a standing self-declaration, confirmed annually by the institution itself, and case stories are institution-authored and listed simply as examples of the guide in use.

6. Adoption Pathway and the Prediction–Validation Loop

GHP's unit of adoption is the module: an institution may declare GHP-100 alone and be listed as a GHP-100 adopter. GHP-600 follows the same one-module-at-a-time logic. Its obligations become effective when an institution offers interpretation (GHP-300) or review (GHP-500) services; adopters of inquiry-only modules may defer them. This keeps the entry barrier low without diluting the discipline where it matters most, which is wherever causal claims are made. The registry is honesty-gated: it lists only real declarers, and stays visibly empty until the first declaration is confirmed.

Every adopted closed loop produces a structured record: registered hypotheses and windows on one side, re-test outcomes on the other (Figure 5). Aggregated across adopters, these records form a shared body of prediction-and-outcome evidence that calibrates the guide itself: which measurements carry the most information, which phenotypes are hardest to trace, and which lag windows mis-specify. The guide thus becomes a learning system in the same sense it demands of its adopters [10]. How such individualized records aggregate across very different adopters, without flattening the individual differences that make them valuable, is a genuinely open question; the corpus is built to expose it rather than assume an answer. Data returned by adopters feeds under a consent-and-license framework, with the corpus treated as a trade-secret-protected asset of the initiative.

7. Relation to Prior Standards and Literature

GHP inherits the GxP pattern: practice codes issued when an activity becomes trust-critical [1]. It extends that pattern with a quality discipline that process codes historically lacked: the quality of the reasoning itself. The process track parallels international service-quality standards but grounds them in causal substance: four-layer inquiry, root-cause positioning and contribution review. The second track draws on four literatures. N-of-1 trial methodology treats the individual as their own control [9,15]. Real-world-evidence frameworks achieve rigor by design and documentation rather than by setting [11]. Preregistration culture, imported from clinical trials, disciplines individual care plans [16]. Learning health systems close the loop between practice and knowledge [10]. Within our own body of work, the root-cause requirement (GHP-300) rests on a layered architecture that traces exposures to root causes to functions to phenotypes. GHP-500 and criterion 600.7 rest on a methylome-scale audit asking when multi-intervention effects can be decomposed at all. The reasoning behind GHP-600 draws on our own work on steerable world models and N-of-1 intervention reasoning [17,18,19], and on the emerging science of biomedical world models at large [6]. GHP is meant to be an international, data-driven initiative built on these two lines: the N-of-1 line disciplines how one person's evidence is generated, and the world-model line gives today's records a future use. Counterfactual reasoning is the stone both lines stand on. Steerability, meaning the ability to ask a model what would happen if we intervened here instead and get an auditable answer, is what turns records into decisions [17].

8. Outlook: What Today's Records Become

For a longevity clinic, the most consequential change on the horizon is not a new assay or a new supplement. It is the arrival of models that can predict, before a treatment is given, how a specific person will respond. Research on such models now spans several biological levels [6], from single cells [20] to lab-grown tissue [21] to whole individuals [17,18,19]; for a clinic, only the last level matters, and it is usually called a virtual patient.

What do virtual patients have to do with a practice guide? Everything, because models of this kind are trained on exactly one thing: records of what was done to real people and what happened next. A clinic that follows GHP produces such records as a by-product of honest practice. Today those records discipline the clinic itself; accumulated across many adopters, they become the raw material from which the next generation of virtual patients will learn. One early instance from our own work reasons about N-of-1 interventions in a single person [18]. Early adopters are not merely complying with a code; they are compiling the corpus the future will be built on.

Two names describe that future. The raw material is intervention-response data: the accumulated records of which intervention moved which marker, in whom, and by how much. The goal such data makes possible we call capomics: for each individual, a measurable profile of what his or her biology can and cannot respond to. GHP is the bridge from today's practice to that goal, and its own criteria (600.1, 600.2, 600.7) already require the discipline the transition will demand.

An open ecosystem around a short shared list. GHP is built as an open interface, not a closed pipeline. Assay and laboratory vendors meet it wherever a molecular aging anchor is needed (GHP-200); data platforms and AI developers meet it in the prediction–validation records that every adopted loop produces; model builders meet it in the corpus those records accumulate into. What keeps this openness from dissolving into noise is the short shared list at the center: the hallmarks of aging, a common coordinate that clinics, vendors, laboratories and models can all organize around. The guide is meant to grow with the world models it feeds: as biomedical world models get better at counterfactual and steerable reasoning [6,17], the guide's thresholds and windows will be re-fixed against that capability, version by version.

GHP v0.2 remains a draft, and honestly so. Several numerical parameters are deliberately left open until platform data can fix them; the guide's effect on practice and client outcomes has not yet been evaluated; and the founding endorsement list is still in invitation. The draft is modular precisely so that the community can test, criticize and improve it module by module. We plan to evaluate the guide empirically as adoption accumulates, and to report the results under the same honesty rules the guide imposes on others. Public comment is open through 2026-12-31. We publish now, before everything is resolved, because a guide that waits for perfect evidence cannot shape how that evidence is generated.

The broader claim is simple. In longevity medicine the scarce resource is not enthusiasm, not assays, not even trials. It is trustworthy individualized inference. GHP is an attempt to make that inference a public, auditable craft: trace the cause, measure at the molecular level, read the story, treat the root, prove what worked, infer with integrity.