ELITE: Deep Profiling and Algorithmic Polytherapy to Map Individual Response in Healthy Aging

Study purpose and design ELITE, Evidence-Based Longevity Interventions and Targeted Engagement, was a randomized, open-label, two-active-arm Phase IIa pilot conducted at Sheba Medical Center in Ramat Gan, Israel. The study was designed to characterize individual response to multidomain healthy-aging interventions by combining randomized treatment with frequent digital monitoring, scheduled clinical and functional assessment, multi-omic profiling and a prespecified Bayesian response framework.

The central scientific question was whether change during treatment could be followed at higher temporal resolution than is possible with clinic visits alone, and whether those longitudinal digital trajectories could be related to measured changes in muscular, immune and cognitive function. Treatment exposure, digital measurements, clinical assessments and molecular data were therefore organized on the same participant timeline.

ELITE evaluated a protocol-defined algorithmic polytherapy intervention that integrated multicomponent treatment exposure, high-frequency longitudinal digital measurement, a prespecified Composite Response Score (CRS/ADAPT Score), Bayesian response-state classification, and physician-supervised clinical review. The treatment component consisted of randomized metabolic-drug regimens together with shared AgeBiotics AB-01 and AB-02. The digital and computational components were part of the same intervention architecture, providing longitudinal response monitoring and decision support throughout the study. During this fixed-window pilot, randomized arm allocation remained unchanged; KEEP, AUGMENT, SWITCH, and STOPP states were generated for physician review but did not trigger algorithm-guided reassignment.

ELITE was developed in connection with Ani Biome's (developer of the ANI.AI platform) participation in the XPRIZE Healthspan competition. Muscular, immune and cognitive function were incorporated into the Muscular-Immune-Cognitive Composite (MIC-C), the study's principal multidomain clinical measure of change from baseline to Week 8.

Study population and setting Thirty-one community-dwelling adults aged 50 to 80 years without diabetes and with a body-mass index of at least 25 kg/m2 were randomized in a 1:1 ratio. Screening HbA1c was below 6.5%. Participants were able to provide informed consent, attend scheduled study visits and complete app-guided home assessments. The study was conducted in a healthy-aging population meeting these protocol-defined eligibility criteria.

Participants were recruited primarily from the Sheba Healthspan Research Population (SHARP), with additional recruitment through community and clinical-network outreach. Scheduled assessments were performed at Sheba Medical Center, while frequent longitudinal digital measurements were completed in participants' daily environments through the ANI.AI platform.

Randomized multidomain interventions Participants were assigned to one of two active regimens for approximately eight weeks. Arm A received metformin 425 mg orally once daily plus semaglutide 0.25 mg by subcutaneous injection once weekly. Arm B received metformin 425 mg orally once daily plus empagliflozin 10 mg orally once daily.

Both groups also received AgeBiotics AB-01 oral solution, administered as three 19-mL sachets per day, and AB-02, administered as five capsules twice daily. Because AB-01 and AB-02 were given in both arms, the randomized comparison was defined by the metformin-plus-semaglutide and metformin-plus-empagliflozin regimens. Digital, functional and molecular changes were followed within each randomized arm during treatment.

Medication reminders and intake confirmation were incorporated into the ANI.AI workflow. Pharmacy and study records documented dispensing, dosing, interruptions and discontinuations. Randomized arm allocation remained fixed throughout the intervention, while individual study products could be interrupted or discontinued for clinical or safety reasons under investigator oversight.

Hybrid clinic-home measurement ELITE combined scheduled clinical visits with high-frequency app-guided monitoring at home. Clinic visits provided standardized clinical, functional, laboratory and molecular assessments. Between visits, participants completed structured digital sessions through ANI.AI, allowing day-to-day changes to be examined alongside the less frequent clinic-based measurements.

Digital observations, treatment records, clinical tests, laboratory values, molecular samples, adverse events and computational outputs were linked by participant and date. This made it possible to compare overall baseline-to-Week 8 change with the sequence of measurements collected during the intervention.

High-frequency longitudinal digital phenotyping Participants completed guided digital assessment sessions through the ANI.AI application. Each session included tongue, left-eye, right-eye, face and wrist images, a structured micro-survey and a voice recording. A dedicated capture device was used for multispectral measurements of the tongue, face and wrist.

Each digital session was timestamped and linked to its participant, modality and source record. This created a traceable longitudinal data layer connecting the original observation to derived digital features, computational outputs and response estimates.

Repeated digital measurement provided temporal resolution between clinic visits. It enabled characterization of response onset, direction, rate of change, persistence and within-person variability, and supported analyses linking early or evolving digital patterns with later clinical and molecular measurements.

The digital workflow was also a core feasibility component. ELITE evaluated adherence to repeated app-guided assessments, completeness of required modalities, processing success and the operational integration of daily digital measurement with active treatment. Digital observations collected after the required intervention window provided an additional longitudinal layer for evaluating persistence of participant trajectories.

Clinical and functional phenotyping Scheduled study visits included broad clinical and functional phenotyping across multiple physiological systems. Assessments included body composition, cardiorespiratory and pulmonary function, vascular and autonomic measures, electrocardiographic measures, muscular performance, mobility and frailty, cognition, psychological and sleep-related measures, endocrine markers and clinical laboratory testing.

Muscular, immune and cognitive measures were integrated into the MIC-C. The primary protocol outcome was change in MIC-C from baseline to Week 8, with muscular, immune and cognitive subdomains evaluated separately as secondary outcomes. These clinical and functional measurements also served as reference points for interpreting the higher-frequency digital trajectories.

Multi-omic and molecular profiling Blood, urine, stool and saliva samples were collected at scheduled study time points for deep molecular profiling. Samples were collected for genomic, epigenomic, transcriptomic, proteomic, metabolomic, glycomic, metagenomic, regulatory-RNA and immune profiling, together with additional targeted biological assays.

Assay layers acquired to date include broad metabolomic and proteomic profiling, IgG glycomics, microbiome metagenomics, whole-genome bisulfite sequencing for epigenomic analysis and RNA sequencing for transcriptomic analysis. Molecular data are linked to the longitudinal participant record as assay-specific quality control and integration are completed. The collected biospecimens support a broader multi-omic analysis program.

These multi-omic measurements provide molecular context for the digital and functional trajectories and support analyses of biological pathways associated with treatment response, heterogeneity of response and participant-specific state change.

Longitudinal data integration ELITE linked baseline measurements, treatment exposure, repeated digital observations and scheduled clinical and molecular assessments by participant and time. This created a single longitudinal record for examining how measured changes unfolded during treatment.

Source information was retained for acquired measurements and the features and model outputs derived from them. Computational response estimates could therefore be traced back to the digital observations, clinical measurements and treatment history that informed them.

The combined dataset supports analyses of digital-clinical and digital-molecular relationships, prediction of later clinical measurements from earlier observations, trajectory slope and variability, and factors associated with differences in response between participants.

Composite Response Score and Bayesian response states The Composite Response Score (CRS/ADAPT Score) is a high-frequency representation of within-person change derived from longitudinal digital biomarker data. It summarizes evolving response relative to each participant's own baseline and is updated as new participant-specific evidence accumulates.

The ANI.AI response framework used Bayesian posterior probabilities to translate longitudinal evidence into four prespecified response states: KEEP, AUGMENT, SWITCH and STOPP. These states provided a consistent language for likely response, uncertainty, nonresponse and potential harm and were reviewed within the physician-supervised clinical workflow.

The CRS and the Week 8 MIC-C operate at complementary timescales. CRS provides a high-frequency digital response trajectory updated over time during treatment, while MIC-C provides a scheduled multidomain clinical assessment across muscular, immune and cognitive function. The protocol prespecified evaluation of the CRS-driven response framework against the Week 8 MIC-C responder criterion, including discrimination using area under the receiver operating characteristic curve.

CRS trajectories, posterior probabilities and response states were generated during the pilot and retained with their associated timestamps and source data. These records support the prespecified evaluation of whether the high-frequency digital response trajectory identifies participants who meet the Week 8 MIC-C responder criterion.

Longitudinal digital health composite and modeling The study included a complementary longitudinal digital health composite derived from repeated digital biomarkers and outputs of domain-specific computational models. It characterizes multidomain change across the intervention period and provides a broader longitudinal representation of participant trajectory alongside the response-focused CRS.

Planned analyses include the association between change in the longitudinal digital health composite and change in MIC-C, together with trajectory properties such as slope and within-participant variability. These analyses connect high-frequency digital change with measured clinical change and support development of participant-specific longitudinal models.

Participant-specific longitudinal modeling The ELITE dataset links repeated digital measurements with treatment exposure, clinical and functional assessments, and multi-omic data. Future work will use these linked observations to develop participant-specific models and test whether they can reconstruct measured biological and functional properties, forecast later outcomes and characterize individual treatment response.

Clinical oversight Clinical oversight included physician review, adverse-event assessment, product accountability, source verification and external monitoring. The Sheba Helsinki Committee / Institutional Review Board approved ELITE (2361-25-SMC), with informed consent from all participants. Sheba's Sponsor's Office monitored the study, and The Sheba Fund for Health Services and Research was regulatory sponsor of record.