Buck Institute and Oracle Health Plan a Score for Age-Related Decline

On October 30, 2026, NutraIngredients reported that the Buck Institute for Research on Aging and Oracle Health had partnered to study biomarkers of age-related decline. The collaboration will use de-identified longitudinal health records from Oracle’s Oracle Life Sciences Data Intelligence platform.

Why intrinsic capacity is the focus

The project’s central aim is to develop and statistically validate an Intrinsic Capacity Score that identifies signals associated with frailty or chronic-disease risk before symptoms appear. For healthspan research, the relevant question is whether information recorded earlier in life can help anticipate later decline. NutraIngredients describes an effort to connect those earlier signals with meaningful health outcomes. Its significance lies in that research objective: developing a measure that could eventually support the study of aging and decisions about care.

How the project will use human health records

Research setting: planned analysis of human health records, not in vitro experiments or animal models.

According to NutraIngredients, Buck Institute researchers are expected to receive secure, hosted access to de-identified real-world data. They will be able to build cohorts, run queries and conduct bioinformatics workflows. These capabilities describe how researchers will work with the records and organize the analyses needed to develop the score.

The records are longitudinal, meaning they contain health information collected over time. That feature fits the project’s stated goal of looking for signals associated with later outcomes. The central research question concerns the relationship between earlier information and subsequent health, rather than a measurement taken at an isolated moment.

The proposed score addresses five intrinsic-capacity domains. The report describes cognition and locomotion alongside sensory function, vitality and psychology. This breadth is important to the project’s stated purpose: assessing decline across several aspects of function rather than centering the entire effort on a single measurement.

David Furman directs the Buck Institute’s AI and Bioinformatics Platform. He told NutraIngredients that building predictors from large real-world datasets will require regularization methods that automatically select features carrying the most signal. The report also describes planned automated feature selection to identify informative predictors from a large set of variables.

The useful distinction here is between access to information and the selection of information that contributes to a prediction. The reported analytical plan addresses the latter task. For readers assessing the eventual model, the question will be how the selected predictors connect to the outcomes the score is intended to forecast.

The collaboration also belongs to a broader research effort. NutraIngredients places the work within THRIVE, short for Transforming Health: Reclaiming Intrinsic Vitality for Everyone. THRIVE is a Stanford-led consortium within ARPA-H’s PROSPR program and includes the Buck Institute.

According to the report, the coalition received up to $34.5 million from ARPA-H. That funding belongs to the consortium effort rather than solely to the Buck Institute and Oracle Health partnership. Keeping the organizational scope clear helps distinguish the resources supporting the wider program from the specific data collaboration described in this announcement.

Together, these details establish the project’s research setting: human records, a planned modeling approach and a consortium-supported objective. A useful reading frame is the connection between aging measurements and meaningful health outcomes. That connection is the central question for interpreting the proposed score.

Why the evidence needs closer scrutiny

The announcement describes a development program rather than a completed study with a primary finding. Its main limitations also provide a practical framework for assessing future reports:

• Platform size is not the study sample. NutraIngredients reports that Oracle’s platform contains more than 122 million de-identified real-world data records. The article does not state how many will be eligible for this project or included in its analyses. Future reporting should distinguish the platform’s total holdings from the records used to develop the score and those used to validate it.
• Predictive performance has not been reported. The article provides no sensitivity, specificity or calibration results. It also supplies no external-validation results. Readers should look for published performance measures tied to a clearly described validation population rather than treating access to a large dataset as evidence of predictive accuracy.
• The underlying population is not characterized. The report does not provide details about who is represented in the records or how the team will address potential bias. De-identification describes a feature of the data but does not answer questions about population coverage. A future validation report should explain the population studied so readers can judge the relevance of its findings to other groups.
• Missing data and follow-up remain unspecified. The article does not explain the extent of missing information or the follow-up available for the proposed analyses. These are important questions for a project linking earlier records with later outcomes. When results appear, readers should look for a clear account of which observations were usable and how incomplete records were handled.
• The forecast horizon is an ambition. NutraIngredients describes a consortium goal of forecasting major health outcomes up to 20 years ahead. That is not an established prediction window for an individual patient. Future claims should pair each stated horizon with the outcome being forecast and the validation evidence supporting that particular use.
• The intended outcomes need precise definitions. The report identifies frailty and chronic-disease risk as targets while describing a score spanning multiple functional domains. To interpret later findings, readers will need clear definitions of the outcomes used to judge success. The practical question is not simply whether a score changes but whether that change has a demonstrated relationship to the health outcome under discussion.
• Intervention measurement is a separate research goal. Furman described a possible future use of the score for measuring the effects of lifestyle or dietary interventions on age-related decline. The announcement does not demonstrate that capability. When evaluating such claims, look for evidence that directly tests the score in the proposed intervention setting rather than assuming a risk-prediction result establishes that additional use.
• An at-home assessment is conditional. Furman said a longer-term aim could be a kit combining blood biomarkers with functional data and a questionnaire if the model is successfully validated. This was not an announcement of an available consumer or clinical test. Readers should treat the kit as a possible later application and keep personal screening decisions grounded in qualified care rather than this research announcement.
• Regulatory qualification has not been established. The report describes FDA qualification as a possible objective rather than an achieved milestone. Statistical validation and a stated regulatory ambition should therefore be read as distinct parts of the proposed development path. Any future claim of qualification should be assessed on its own documented evidence rather than inferred from the partnership or its funding.

The practical approach is to evaluate each future claim against the evidence needed for that specific use. A forecast of later frailty, a measure of intervention response and a home assessment are different proposed applications. This framework for judging biological-age models offers a related way to keep the intended use central when reading about an emerging measurement.

How the next phase should test the score

The next decisive step is a published validation study that tests the proposed score against clearly defined health outcomes in a separate population and reports performance across the intended prediction period.

How AgeAmaze helps

Frailty forecasts need a clear connection to later health outcomes, and AgeAmaze makes the science behind that connection readable. For adults sorting human-record research from animal experiments, the publication distinguishes evidence types and explains uncertainty around emerging biomarkers so the research setting remains part of how a claim is judged.

Read the research

Sources

1. Buck Institute, Oracle partner to detect age-related decline years before symptoms