I looked for every health data system that shares features with my Medical Snapshot proposal, something I’ve been working on since 2007. Such programmes medically measure people over time, store what they find, and don’t usually tell the individuals what their own data says. This article summarises what I have learned by looking at 25 existing systems, covering 90 years from the Tuskegee Syphilis Study ↗ (1932) to Our Future Health ↗ (2022). I have tried to indicate my biases and assumptions but the purpose of these notes is to inform my own thoughts about the Snapshot concept.
What I measured
I scored each system against 18 features of the Snapshot. UK Biobank, for example, scores 7.5 out of 18 and yet is one of the foremost voluntary biobanks in the world, so clearly my review is from one very narrow viewpoint!
This article is about the process of generating the full scoring table, which is as exhaustive as it is boring. I hope it will help in some future experimental study design.
Headline results
Plot 1: Research mechanics (x) vs Snapshot participant-benefit features (y, max 6). Research biobanks cluster at moderate mechanics and about 2 of the 6 participant-benefit features, regardless of country or scale. Above that cluster the plot is empty apart from the Snapshot.
Plot 2: Research mechanics (x) vs Snapshot governance features (y, max 11). The bracket marks eight governance features that no operational system has combined with high mechanical capability.
The R source for both plots, with data embedded goes with the scoring page.
The Western forensic databases (NDNAD ↗, CODIS ↗) score 3/18, the same as the China MPS DNA database ↗. They share the same pattern: biological banking, no individual feedback, clinical care independence, with inversions on population type, state control, and voluntariness. It might seem strange to Western and Eastern sensibilities to give them the same score, but such sensibilities are not the point of this review.
Searches and biases
The indexed English-language bioethics literature is large and easy to search. My comparison started there with Framingham ↗, UK Biobank ↗, deCODE ↗, and FinnGen ↗. We always expect bias, but this corpus is very evidently focussed on finding the answers they expect to the questions they have asked, and quite lacking in curiosity. I tried some translations of English queries into various languages, but various built-in biases meant that there were very few results because other scientific traditions view this problem very differently. As an example: translating (via both machine and obliging native chinese speaker) the phrase “biobank observer study no feedback ethics” into Chinese produces work responding to English-framed questions. That is interesting but completely unhelpful.
Therefore I made subsequent searches in Chinese, French, Spanish, and Russian using native-language framing, which seemed to address the bias inherent in translated English queries. Thus, searching for 健康医疗数据相关研究的伦理审查 (“ethics review of health data research”) reaches the PRC legal and health-policy literature on its own terms, and returns a wealth of discussion.
In a similar vein, the Francophone cohort governance literature is almost invisible in the English-language sources despite being quite despite being the most developed framework I found anywhere for healthy volunteer ethics. This is the INSERM VolREthics charter ↗ and a two-volume IGAS report on cohort studies ↗. This has substantially formed my thinking on this general topic, because it is the collective wisdom of dozens of specialists over many years on this very topic. The Spanish-language public health literature gave me Gaceta Sanitaria ↗ on big-data health research ethics.
The Russian-language sources I reached were mostly regulatory and compliance-focused, with little of the argument and dissent the other four turned up, which is itself a result.
What the Scores and Plots Show
The pattern is consistent across both plots. Research biobanks cluster at mechanics 3–5.5 and two participant-benefit features, regardless of where they are or how large they are. Framingham ↗ (1948, USA), UK Biobank ↗ (2006, UK), and the China Kadoorie Biobank ↗ (2004, China) sit in the same cluster despite being separated by six decades and radically different governance contexts. So I feel that this viewpoint has been quite useful despite its obvious problems.
The DoD Serum Repository ↗ scores highest among operational systems (8/18). It predates the Snapshot proposal by 22 years and is equivalent in many respects. In terms of my critera, it fails on voluntariness (compulsory military), state exclusion, benefit-sharing, commercial funding, and public release.
The novelty of the snapshot system is where there are sparse or empty columns. These are column C (retroactive release), D (directionality), C9 (clinically triggered release) and D11 (participant can initiate release).
Generation Scotland ↗ returns basic clinical measurements to participants and, with permission, to their GP, which is why it gets 5.5/18.
Where Assumptions Are Embedded
Dimensions A and B are fairly neutral towards the point of view of any assessor, including me. Whether a system tests healthy people repeatedly, retains samples, or withholds results is are structural facts.
Dimensions D, E, and F imply reviewer assumptions, and my assumptions tend to the Western idea of rights-based traditions:
D13 (voluntary consent) is a binary choice to join or not. In contexts where community obligation rather than individual choice governs participation norms (Confucian healthcare settings, some African community consent frameworks), much more nuance is needed. The [C] notation suggests where these considerations may apply, but it’s only a hint that we are out of our depth and specific knowledge is needed.
E14 (benefit-sharing) assumes that the Snapshot’s idea of payment is the ideal, for which there is no evidence but it does support the point of view of entire review. This view agrees with the French healthy-volunteer tradition set out in the Global Ethics Charter for the Protection of Healthy Volunteers ↗. On the other hand it is regarded as extractive and exploitative in the African tradition, where payment above expense reimbursement may compromise the liberty to volunteer.
Majengo ↗ funds around US$1,088 of healthcare per woman per
year but I scored it as 0 because in-kind community benefit is a different thing from
wages. All of Us ↗ pays US$25 for an in-person visit and scores 1. A reasonable
person might decide that Majengo shares more with its participants than All of Us
does. I found a helpful scoping review of benefit-sharing in biobanking ↗ that shows
disagreement over what benefits should count and who should receive them.
Fifteen of the 25 systems are marked [?] because I reached my personal limit of what I could check.
F17 (governance accountable to participants) embeds English trust law. The concept exists in some form in most legal systems, but the specific model the Snapshot assumes is not universal.
Other designs emerging from other traditions might replace payment with community benefit-sharing, or replace trustee governance with state oversight backed by strong individual rights provisions. My review doesn’t attempt to record this, only differences relative to the Snapshot system.