Estimated avoidable health cost (illustrative)
–
Hazards, highest priority first
| HHSRS hazard | Risk | Why |
|---|
Recommended interventions, best value first
| Intervention | Cost | Score + | Benefit / yr | Payback |
|---|
Score distribution, 1,000 synthetic homes
Average score by neighbourhood (synthetic homes, real ward names; axis does not start at 0)
Budget planner
Greedy allocation by estimated health benefit per pound, at most one measure per hazard per home. Illustrative only.
Measures funded
| Measure | Homes | Cost | Benefit / yr |
|---|
Fix-first list
| Home | Area | Type | EPC | Score | Top hazard | Est. cost / yr |
|---|
Automated validation
This runs randomly generated synthetic scenarios against the model's safety and logic rules (its invariants) in your browser. A TRL 3 proof of concept should show that the core function behaves as specified.
Not run yet.
| # | Rule (invariant) | Checks | Failures | Result |
|---|
How the score works
- Likelihood (0–1) for six HHSRS hazards, from property features (EPC band, heating, glazing, damp, stairs, alarms, crowding).
- Vulnerability multiplier (≥1) from the household, following the HHSRS most-vulnerable age groups (e.g. 65+ for excess cold, under 14 for damp and mould, 60+ for falls) plus simple health flags.
- Weighted risk: each hazard's weight reflects its relative share of NHS cost in BRE's national estimates, where excess cold is the largest.
- Healthy Home Score = 100 × (1 − weighted normalised risk). 100 means no modelled hazard.
- Interventions are re-scored by applying them to the home. Benefit is the fall in estimated annual health cost, and value is benefit ÷ cost.
Is the validation itself trustworthy? Three deliberately broken versions of the model (damp scale reversed, older age lowering cold risk, a repair that makes the home worse) were each caught by the validation rules.
Every number is visible and explainable. There is no machine learning at this stage, by design.
Parameters (v)
| Hazard | Weight | Illustrative cost if realised / yr |
|---|
All parameters are placeholders for calibration against the BRE Housing Health Cost Calculator, local housing condition data and linked primary care outcomes (TRL 4).
Technology readiness
| TRL 2 | Concept formulated: housing condition × household vulnerability → health risk → prioritised action. | Done |
| TRL 3 | Proof of concept: working scoring model, intervention ranking, portfolio optimiser and automated validation on synthetic data (this Lab). | Demonstrated |
| TRL 4 | Validate in the lab on real but de-identified housing stock data. Calibrate weights with a health economist and statistician. | Next |
| TRL 5–6 | Pilot with a housing partner and H&B practices in Wolverhampton under DPIA, DCB0129/0160 and DTAC. | Planned |
Evidence base
- BRE (2021) The cost of poor housing in England: poor housing costs the NHS about £1.4bn a year. Excess cold accounts for about £857m of this, and 2.6 million homes (about 11%) had a Category 1 hazard in 2018.
- Housing Health and Safety Rating System (HHSRS): 29 hazards, scored by likelihood and harm, with Category 1 for scores of 1,000 or more.
Governance built in
- Synthetic data only. No cookies, analytics or server calls.
- Proposes, never decides. A person makes every decision.
- Before any real data: DPIA, data-sharing agreements, Caldicott approval, DCB0129/0160 clinical safety case, DTAC and penetration testing.
- A medical device regulation (MHRA) assessment is required if outputs are used to guide an individual's care.
Founders: Dr Saini (Lead) and Dr Rajnish Mohindroo. In development with Health and Beyond, Wolverhampton.