The conversion layer. Baseline Risk and the Relative-Absolute Split holds the machinery and, as of today, study-control-rate baselines; this supplies the missing piece — a prognostic baseline for a named stratum rather than a trial population’s event rate. Every relative effect in the corpus becomes an absolute one once a stratum-specific 10-year risk exists.
The thresholds — and they move with age
| <50 years | 50-69 years | >=70 years | |
|---|---|---|---|
| Low-to-moderate: “risk factor treatment generally not recommended” | <2.5% | <5% | <7.5% |
| High: “risk factor treatment should be considered” | 2.5 to <7.5% | 5 to <10% | 7.5 to <15% |
| Very high: “risk factor treatment generally recommended” | >=7.5% | >=10% | >=15% |
[EXTRACTED (ESC - CVD Prevention Guidelines 2021) Table 5]
The same 10-year risk can mean two different things at two ages — 8% is very high under 50 and high at 72. It cannot mean three — the thresholds are monotone, so no single number can be very-high in one band and low-to-moderate in another. The stated reason for the banding: “Age is the major driver of CVD risk”, so a fixed threshold would over-treat the old and under-treat the young.
Two qualifiers ESC attaches itself:
- At >=70, the lipid-lowering recommendation drops to Class IIb (“may be considered”) even in the very-high band.
- The banding is an artefact of presentation: “age is obviously continuous, and a sensible application of the thresholds in clinical practice would require some flexibility… as patients move towards the next age group, or recently passed the age cut-off.”
Read-off baseline risk — moderate-risk region, men, ages 60-64
10-year risk of fatal and non-fatal CVD, in percent. Columns are non-HDL cholesterol (3.0-3.9 / 4.0-4.9 / 5.0-5.9 / 6.0-6.9 mmol/L); rows are systolic BP.
| SBP | Non-smoking | Smoking |
|---|---|---|
| 160-179 | 11 · 12 · 13 · 15 | 17 · 18 · 20 · 22 |
| 140-159 | 9 · 10 · 11 · 12 | 14 · 15 · 17 · 18 |
| 120-139 | 7 · 8 · 9 · 10 | 11 · 13 · 14 · 15 |
| 100-119 | 6 · 7 · 7 · 8 | 9 · 10 · 11 · 12 |
[EXTRACTED (ESC - CVD Prevention Guidelines 2021) Figure 3, moderate-risk region chart, read directly
from the rendered page per CLAUDE.source-prep.md -> Tables, tier 3]
Moderate-risk region = Austria, Cyprus, Finland, Germany, Greece, Iceland, Ireland, Italy, Malta, Portugal, San Marino, Slovenia, Sweden. Region is not a detail — the four charts differ substantially for identical inputs. The size of that regional gap is deliberately not quantified here: this guideline does not state it, and the figures that circulate for it come from the SCORE2 algorithm paper, which is staged but not ingested.
A stratum defined by age and condition cannot be read off this chart. Layer 1 - Ranking Interventions for a Stratum fixes age, sex and condition; the chart also needs smoking status, a BP band, a cholesterol band and a region. Those four are case inputs supplied at layer 3, not properties of a stratum — and the answer moves with them:
| Assumed inputs | Read-off | ESC category (50-69) |
|---|---|---|
| non-smoking, SBP 140-159, non-HDL 4.0-5.9, moderate region | 10-11% | very high — treatment generally recommended |
| same but SBP 120-139 | 8-9% | high — treatment should be considered |
| same but smoking | 15-17% | very high |
So the ESC category flips on inputs no stratum contains. This is the layer-2/layer-3 boundary made concrete: the fabric can hold the grid and the thresholds, and it cannot produce a category without a case. Three rows are shown rather than one deliberately — selecting a single read-off would smuggle in assumed inputs and manufacture a decision-change out of the writer’s own defaults.
The conversion, worked once, with its assumptions visible
The wiki holds Saturated Fat Intake and Replacement: reducing SFA gives RR 0.83 (0.70-0.98) for cardiovascular disease. At a 10% baseline:
absolute reduction = 10% x (1 - 0.83) = 1.7 percentage points, or ~17 fewer CVD events per 1000 over 10 years — with the interval running ~2 to 30 fewer per 1000.
Three assumptions, none of them free, and all of them the reader’s to accept or reject:
- The relative effect transfers to a 10-year horizon. The trials behind RR 0.83 ran for varying, mostly shorter periods; SCORE2’s baseline is explicitly 10-year. Constant-RR-over-time is an assumption, not a finding.
- The outcome definitions match. SCORE2’s endpoint is fatal and non-fatal CVD; WHO’s composite is “cardiovascular diseases” from its own reviews. Similar, not verified identical.
- The trial populations transport to this stratum. SCORE2 excludes people with diabetes from the base model — and this stratum has metabolic syndrome, which sits adjacent to that exclusion. SCORE2-Diabetes exists precisely because the base model does not cover diabetes, and it is staged but not ingested.
[INFERRED (ESC - CVD Prevention Guidelines 2021; WHO - Saturated and Trans Fatty Acid Intake 2023) — arithmetic on the two held quantities.]
The novelty here is narrow, and the tempting larger claim is false. It is not the case that neither source performs this conversion: WHO performs it throughout — its evidence profiles carry an “Absolute - per 1000” column and report 15 fewer per 1000 (25 fewer to 2 fewer) for this exact RR, at its trials’ 8.5% control rate, using the formula already recorded on Baseline Risk and the Relative-Absolute Split. What the wiki adds is re-baselining to a stratum-specific prognostic estimate rather than a study control rate — a real move, and a much narrower one.
Note also that WHO grades this estimate Moderate certainty; the RR should not travel without it.
What this licenses, and what it does not
- It converts ratios into absolutes — the single missing input on Baseline Risk and the Relative-Absolute Split and the reason Layer-1 ranking was impossible.
- It does NOT make the corpus’s ratios commensurable with each other. Different outcomes still differ: a CVD-event ratio and an HbA1c mean difference do not become comparable by sharing a baseline.
- It is a prognostic instrument, not evidence of effect. It answers what is this person’s risk, never what does this exposure do. Multiplying one by the other is the wiki’s move, not either source’s.
- A prognostic number is a starting estimate, not the end of the chain. Risk Modifiers - When Extra Information Changes a Risk Estimate can move it up or down near a threshold (a zero CAC lowers it); and where the 10-year window understates the case for a younger person, Lifetime Benefit - The Frame for Younger Adults reframes it in CVD-free years gained.
- The threshold is the decision, not the percentage. SCORE2 outputs a number; Table 5 is what turns a number into an action — and it is the part most often dropped when the model is quoted.
Limits
- Two ESC conditions the charts carry and this page must not drop: they apply only to people whose risk factors are untreated or have been stable for several years — often false in metabolic syndrome, where BP and lipids are frequently treated — and “Risk estimates then need to be adjusted upwards as the person approaches the next age category.”
- Region-calibrated, and the region does most of the work at the margin. A risk read from the wrong chart is wrong by more than most interventions move.
- Base model excludes diabetes, which bounds its use for exactly the stratum the wiki is targeting. AWAITS SCORE2-Diabetes 2023 (staged).
- Charts read at tier 3 (direct rendered-page read) because
find_tables()returns nothing usable on this layout. Values above are transcribed from the rendered grid; the full four-region chart set is ~2 560 cells and only the moderate-region men’s 60-64 block is carried here. - No calibration or discrimination statistics are held — those live in the SCORE2 source papers (staged, not ingested). PROBAST+AI does not rank these against each other: it lists calibration, discrimination and net benefit symmetrically, and an apparent emphasis on calibration comes from the title of a paper it cites, not from the tool’s own position.