From glucose logging to Occupational Metabolic Intelligence
The live product solves an immediate problem. The long-term research question is broader.
RoadWell’s live product addresses an immediate problem: making glucose-monitoring routines easier to follow and document around professional driving. But the long-term research question is broader:
Can occupational context help us understand changes in metabolic health earlier and more personally than health measurements viewed in isolation?
Professional drivers do not live under laboratory conditions. Their metabolic patterns can occur alongside changing shifts, long journeys, disrupted meals, sleep patterns and repeated periods of sedentary work. RoadWell is building a consented longitudinal research dataset that links health-monitoring information with appropriate occupational context, to test whether those relationships can generate useful, validated insights.
The Occupational Metabolic Intelligence framework
Layer 1 — Occupational monitoring data
RoadWell currently captures consented monitoring information around the professional-driving workflow, including timestamps, relevant journey context and monitoring adherence.
Layer 2 — Metabolic Shift Signature
A longitudinal representation of how an individual driver’s metabolic pattern changes under recurring occupational conditions — establishing an individual’s own baseline rather than comparing only against a population average.
Layer 3 — Deterioration intelligence
Tests whether occupational signals provide measurable predictive value beyond conventional demographic and health information. Any predictive capability is introduced only after adequate data, validation, clinical governance and appropriate regulatory assessment.
Layer 4 — Adaptive support
Tests whether behaviour support adapted to working pattern, culture and individual response produces better sustained engagement than generic advice.
Layer 5 — Safety constraints
Any future personalised output sits behind deterministic safety, clinical, regulatory and driving-context rules.
How the research system works
RoadWell’s research architecture is designed around longitudinal, driver-level data rather than isolated glucose readings. Where appropriate, consented and permitted, the research dataset may progressively include:
- Physiological context — glucose reading, glucose variability, timing and historical trend.
- Occupational context — shift start and end, shift type, journey duration, time of day and repeated shift patterns.
- Behavioural context — monitoring adherence, meal timing and relevant activity patterns.
- Individual context — the driver’s own historical baseline rather than population averages alone.
Population learning becomes personal learning
A new driver has little personal history, so early research models rely more heavily on population-level patterns. As sufficient high-quality longitudinal observations accumulate, RoadWell can test progressively more personalised modelling.
Evidence before complexity
RoadWell does not introduce a more sophisticated model simply because it is technically possible. Each R&D capability must outperform a simpler comparator against predefined measures.
| Research question | Baseline | What RoadWell tests |
|---|---|---|
| H1 | Memory / paper monitoring | Does a structured digital workflow improve monitoring completeness? |
| H2 | Clinical / demographic information alone | Does occupational context add useful predictive information? |
| H3 | Fixed population thresholds | Can individual longitudinal patterns detect meaningful change earlier? |
| H4 | Static support | Does context-adaptive support improve sustained engagement? |
Model evaluation
Where machine-learning models are evaluated, RoadWell considers discrimination, calibration, sensitivity, false-alert burden and useful lead time. Evaluation is separated at driver level, preventing observations from the same individual appearing across training and test datasets.
Safety is an architectural layer, not a disclaimer
Future personalised systems will not be permitted to send unconstrained model output directly to a driver. Outputs must pass defined confidence thresholds and applicable clinical, regulatory and driving-context constraints before reaching the user.
RoadWell’s design principles include:
- No medication adjustment recommendations.
- No automated licence decision.
- No replacement of healthcare-professional judgement.
- No driver interaction that encourages unsafe device use while a vehicle is moving.
- Low-confidence output defaults to conservative handling rather than forced personalisation.
- Clinical-safety escalation overrides optimisation logic.