What this is

GBD 2021 stroke estimates for 1990–2021, across 204 countries and 23 risk factors. The headline is not the 11.9 million incident strokes or the 7.3 million deaths; it is the 84.1% of stroke DALYs attributable to 23 modifiable risk factors — and the fact that even with a 38.7% reduction in age-standardised DALY rate since 1990, the absolute burden is climbing because the population is ageing and growing faster than risk-factor control can keep up. Most of the attributable risk is metabolic (68.8%), dominated by hypertension (PAF 56.8%); environmental risks (36.7%, led by ambient PM2.5) are increasingly important. The paper is the authoritative reference number every stroke prevention talk in the next decade will quote, and its 2015-incidence-stagnation finding is the most actionable piece of new data.

1. Shadow Audit

a) “84.1% attributable to 23 risk factors” is a ceiling, not a precise estimate. The 95% UI is 77.8% to 88.8%. The 23 risk factors were the ones with sufficient causal evidence in the GBD framework; sickle cell, HIV, atrial fibrillation, and several medication-class exposures were excluded. The true modifiable fraction is almost certainly higher than 84.1% if you include the omitted factors.

b) The 56.8% PAF for systolic blood pressure depends entirely on the TMREL. Theoretical minimum-risk exposure level for SBP is somewhere in the 110–115 mmHg range. Move the TMREL to 130 mmHg and the PAF drops by 15+ percentage points. The relative ranking (SBP ≫ everything else) is robust; the absolute number is not.

c) The “stagnation from 2015” is partly a COVID artefact. The authors explicitly note that 2019–21 saw reduced acute-stroke hospitalisations in many countries. The 2015 inflection predates COVID, but the COVID dip materially flattens the post-2015 trend line. Anyone citing “incidence is no longer falling” needs to flag this.

d) Age-standardised reductions mask divergent population-level trajectories. DALY rate fell 38.7% but absolute DALYs grew by ~50% (population growth + ageing). The age-standardised improvement hides the fact that for an individual hospital, the patient census is larger, younger on average (more strokes in <55-year-olds in HICs), and disproportionately intracerebral haemorrhage in LMICs.

e) “Age-standardised” is GBD’s standard adjustment — but population denominators in 2021 are post-pandemic and migration patterns in 204 countries are not equally captured. The UI on age-standardised rates is wider than the UIs on absolute counts suggest, because the standardisation itself introduces uncertainty.

2. Inversion Engine

For the opposite conclusion — “stroke prevention is failing globally” — to hold, you would need:

• The 38.7% age-standardised DALY reduction to be a statistical artefact of the standardisation model rather than real epidemiological progress.
• The BMI PAF increase of 88.2% to dominate the next decade’s exposure trajectory such that metabolic PAFs return to 75%+ of total stroke DALYs.
• The 2015 incidence stagnation to extend through the 2020s as a secular reversal rather than a plateau-with-Covid-effect.

Quantification of the inversion threshold: if age-standardised stroke incidence rises by 5% globally between 2021 and 2030 — a plausible outcome if BMI, fasting glucose, and ambient temperature PAFs continue their 1990–2021 trajectories — the framing of “global stroke prevention is succeeding” collapses. The flip point is roughly 2× the current BMI exposure prevalence, given current relative-risk estimates.

Mechanism that breaks first: the BMI PAF acceleration. The 1990→2021 rise of +88.2% reflects both rising prevalence and stable relative risk. If BMI prevalence doubles in the next decade (plausible — global adult obesity has roughly doubled since 1990 and continues at ~1–2%/year), the 4.7% global PAF becomes ~9–10%, and the “stroke prevention is mostly about blood pressure” framing erodes.

3. Second-Order Catalyst

If the GBD 2021 estimates are real (and the methodology is sound), the protocol/ institution/ patient subgroup that moves first is:

• Primary-care cardiovascular risk programs in lower-middle-income countries (specifically South-East Asia, sub-Saharan Africa north, and Central Asia). The PAF for hypertension (56.8% globally) climbs to >60% for intracerebral haemorrhage; LMICs have ≥40% prevalence of uncontrolled hypertension. The Lancet Commission recommendation — population-wide BP screening with task-shifting to nurses — is the lowest-cost, highest-yield intervention.
• WHO HEARTS technical package rollout in LMIC ministries of health. The data tells you where to deploy first: regions with highest age-standardised DALY rates are World Bank low-income (DALY rate ~2.5× the high-income rate) and lower-middle-income.
• Atrial fibrillation screening in LMICs is missing from the GBD framework entirely — first-mover sites that establish AF prevalence-stroke-attributable-fraction data would create a new risk-factor line item in GBD 2025/2027.

Adoption pace: WSO-Lancet Neurology Commission recommendations typically take 5–7 years from paper to guideline incorporation, then another 5–10 years to population-level implementation. Expect first-mover impact by ~2030–2033.

4. Asymmetric Leverage

Small effect with huge denominator (the asymmetric upside):

• Population-wide sodium reduction (e.g. UK-style salt-reduction program): even a 1 g/day population-mean sodium drop cuts stroke incidence ~5% per existing meta-analyses. Apply across 8 billion people → tens of thousands of strokes averted annually for essentially zero per-capita cost.
• Tobacco taxation and cessation: A 10% real-price increase cuts smoking prevalence ~4% in LMICs (where PAF is 13.8%). Apply across 1.3 billion smokers globally.

Large effect with small denominator (the targeted upside):

• Hypertension control in the 30% of adults with undiagnosed/uncontrolled hypertension globally: getting that population to <140/90 mmHg cuts their stroke risk ~35–40% per individual. Smaller denominator (~1.5 billion people), but each person saves a stroke event.

The asymmetric lever here is sodium reduction + hypertension treatment — both are cheap, scalable, and PAF-aligned with the 84.1% modifiable burden.

5. Paradigm Destroyer

a) The “stroke is a disease of the elderly” reflex is dead. Adults <70 years now account for the majority of stroke incidence increases globally, particularly in southeast Asia and east Asia. Risk-factor modification programs targeted at “elderly cardiovascular patients” miss the bulk of the contemporary stroke burden.

b) The “lifestyle, not pollution” framing is dead. Ambient PM2.5 has a 16.6% global PAF — comparable to LDL cholesterol, smoking, and fasting plasma glucose. Climate change and air-pollution policy are stroke prevention policy. The WSO-Lancet Commission recommendation for “legislative regulations and taxation of unhealthy products” should explicitly include fossil-fuel and particulate-matter regulation as a stroke intervention.

c) The “intracerebral haemorrhage is rare in HICs” reflex is dead. The proportion of ICH is twice as high in LMICs as HICs, but in absolute numbers, the case mix in HIC stroke units has shifted — more ICH in younger patients, more subarachnoid haemorrhage attributable to non-optimal ambient temperature. ICH is no longer a side note in stroke prevention talks.

d) The “we have fixed the risk factors” reflex is dead. 84.1% attributable, stagnation since 2015, BMI PAF +88.2%, high ambient temperature PAF +72.4% — the risk-factor pipe is accelerating, not stabilising. A 30-second protocol change: every hypertension clinic visit now flags ambient PM2.5 exposure (mask guidance, indoor-air filtration) as part of the cardiovascular risk conversation.

Two-sentence morning protocol update:

“Blood pressure control remains the single highest-yield stroke intervention, but the second-most-important modifiable risk factor is now ambient air pollution — every chronic-disease visit should screen for and address PM2.5 exposure alongside sodium reduction.”

MVP — Minimum Viable Proof

The minimum numerical statement that would change bedside practice is:

56.8% of stroke DALYs are attributable to systolic blood pressure above TMREL; a sustained 10 mmHg population-mean reduction in SBP would prevent roughly one in seven strokes globally.

That single sentence, if it enters a hypertension guideline preface, shifts clinical inertia from “treat individual patients with BP >140/90” to “shift population distribution by 10 mmHg” — and the policy implications (food reformulation, sodium targets, primary-care BP screening) follow.

Best Combination

GBD 2021 should be read alongside:

1. WHO HEARTS technical package (2023 update) — operationalises the PAFs into a clinic- and community-level intervention package.
2. World Stroke Organization–Lancet Neurology Commission on Stroke (Brainin et al., 2023) — the policy roadmap that GBD 2021 quantifies.
3. Lancet Countdown on health and climate change — the mechanistic link between ambient PM2.5 (16.6% PAF), ambient temperature (+72.4% PAF), and stroke.
4. INTERSTROKE case-control study (O’Donnell et al., 2016; 2023 update) — provides the individual-level relative-risk data that GBD extrapolates to population attributable fractions.

The combo is: GBD gives you the magnitude (84.1%, 56.8%, 16.6%); HEARTS gives you the clinic workflow; the Commission gives you the policy frame; INTERSTROKE gives you the individual relative-risk evidence.

Overvalue Warning

a) Don’t overinterpret the 84.1% as “84.1% preventable by individual lifestyle change.” The modifiable burden is a population-level statistical construct; individual risk-factor reduction has heterogeneous uptake, adherence, and biological response. The 84.1% is the upper bound under counterfactual population exposure at TMREL for all 23 risks simultaneously — not a realistic individual target.

b) Don’t overinterpret the 56.8% as “hypertension explains most strokes.” It’s the PAF for SBP exposure above the theoretical minimum; the residual ~43% is split across 22 other risks, with metabolic clustering (BMI, glucose, LDL, kidney) accounting for substantial overlap. Hypertension control is necessary but not sufficient.

c) Don’t extrapolate the 38.7% age-standardised DALY reduction to future decades. The 2015 stagnation, COVID-19 disruption, rising BMI and ambient temperature PAFs, and shifting age structure all argue against linear extrapolation. The next GBD (2025) will be the first real test of whether the trend is recovering or reversing.