For a person with no established cardiovascular disease, does daily aspirin prevent enough cardiovascular events to justify the bleeding it causes? On the largest trial-level meta-analysis held (13 RCTs, 164 225 participants, 1 050 511 participant-years), the answer is no net benefit: aspirin buys a small, real reduction in cardiovascular events that is matched — and slightly exceeded — by an increase in major bleeding, in absolute terms, in every subgroup examined. This is a pharmacotherapy exposure appraised on its own terms; it is the realistic drug alternative a low-risk person weighs against doing nothing.
Zheng & Roddick conclude the two absolute effects are «of similar magnitude», that cardiovascular benefits were «modest and equally balanced by major bleeding events», and that «the decision to use aspirin for primary prevention may need to be made on an individual basis, accounting for the patient’s risk of bleeding» (Zheng & Roddick, 2019).
The net balance — benefit vs harm, in absolute terms
(Zheng & Roddick, 2019) — overall, across all participants (median 10-year baseline CV risk 10.2%):
- Composite CV outcome (CV death + nonfatal MI + nonfatal stroke): HR 0.89 (95% CrI 0.84-0.94); ARR 0.41% (95% CI 0.23%-0.59%); NNT 241 over ~5 years. Event rates 60.2 vs 65.2 per 10 000 participant-years.
- Major bleeding (study-defined): HR 1.43 (95% CrI 1.30-1.56); ARI 0.47% (95% CI 0.34%-0.62%); NNH 210. Event rates 23.1 vs 16.4 per 10 000 participant-years.
Per 1000 person-years this is ~0.5 fewer composite CV events against ~0.67 more major bleeds (inferred from Zheng & Roddick, 2019) (arithmetic on the extracted per-10 000-py event rates). The bleeding harm is not merely comparable — its absolute size marginally exceeds the CV benefit, and the confidence intervals overlap heavily. There is no all-cause mortality benefit (HR 0.94, 95% CrI 0.88-1.01) and no CV-mortality benefit (HR 0.94, 95% CrI 0.83-1.05); deaths from bleeding were rare, so the two arms are weighed as nonfatal events avoided vs nonfatal bleeds caused, not lives saved vs lives lost.
What moves and what does not
| Outcome | HR (95% CrI) | Absolute (95% CI) | NNT / NNH |
|---|---|---|---|
| Composite CV | 0.89 (0.84-0.94) | ARR 0.41% (0.23-0.59) | NNT 241 |
| Myocardial infarction | 0.85 (0.73-0.99) | ARR 0.28% (0.05-0.47) | NNT 361 |
| Ischemic stroke | 0.81 (0.76-0.87) | ARR 0.19% (0.06-0.30) | NNT 540 |
| All-cause mortality | 0.94 (0.88-1.01) | NS | — |
| CV mortality | 0.94 (0.83-1.05) | NS | — |
| Total stroke | 0.93 (0.86-1.02) | NS | — |
| Major bleeding | 1.43 (1.30-1.56) | ARI 0.47% (0.34-0.62) | NNH 210 |
| Major GI bleeding | 1.56 (1.38-1.78) | ARI 0.30% (0.20-0.41) | NNH 334 |
| Intracranial bleeding | 1.34 (1.14-1.57) | ARI 0.11% (0.04-0.18) | NNH 927 |
The benefit is carried by MI and ischemic stroke; there is no mortality signal. The harm is carried by GI bleeding, with a smaller but serious intracranial-bleeding excess.
Baseline risk does not tip the balance — route (a), not route (b)
The natural stratification hope is give it to higher-risk people, where the benefit is bigger. The data defeat this. The relative CV effect is essentially constant across strata — composite HR 0.87 (low risk), 0.91 (high risk), 0.90 (diabetes), 0.89 (overall) — i.e. no effect modification (route-(b) is negative). Absolute benefit does rise with baseline risk (route-(a)), but so does the absolute bleeding harm, and by the same order. In every stratum the bleeding ARI equals or exceeds the CV ARR (Zheng & Roddick, 2019):
| Stratum (10-yr CV risk) | Composite CV: HR — ARR — NNT | Major bleed: HR — ARI — NNH |
|---|---|---|
| Low (<10%, median 6.8%) | 0.87 (0.79-0.95) — 0.34% (0.14-0.52) | 1.45 (1.28-1.63) — 0.40% (0.25-0.57) — NNH 249 |
| High (>=10%, median 12.8%) | 0.91 (0.84-0.98) — 0.63% (0.18-1.04) — NNT 160 | 1.41 (1.23-1.61) — 0.64% (0.35-0.97) — NNH 152 |
| Diabetes | 0.90 (0.82-1.00) — 0.65% (0.09-1.17) — NNT 153 | 1.29 (1.11-1.51) — 0.80% (0.29-1.39) — NNH 121 |
Reading it: raising baseline risk raises both the benefit and the harm arm together, so the net stays a wash-to-negative — the high-risk and diabetes strata are, if anything, slightly worse net (bleeding ARI 0.64% vs benefit 0.63%; 0.80% vs 0.65%) because bleeding risk co-travels with cardiovascular risk. Diabetes is not a positive effect-modifier: the diabetes composite HR (0.90) matches the overall (0.89), and its net absolute balance is unfavourable — a route-(a) baseline-risk story, carrying no subgroup claim. This is the worked instance of absolute benefit scaling with baseline risk while the relative effect is fixed, with the twist that the competing harm scales too, so stratifying on CV risk alone cannot recover a net benefit -> Baseline Risk and the Relative-Absolute Split.
Data note (inferred from Zheng & Roddick, 2019): the low-risk composite-CV figures above are taken from Figure 2 (ARR 0.34%, HR 0.87). The Results prose for the low-risk stratum repeats the high-risk values verbatim (HR 0.91 / ARR 0.63% / NNT 160) — an apparent transcription duplication; the forest-plot figure is treated as authoritative. The bleeding and secondary numbers are internally consistent across prose and figure.
Studied range, dose, and generalisability
- Baseline risk studied: median 10-year CV risk 10.2%, range 2.6%-30.9% (trial-level). The balance is characterised across this whole span; extrapolation above ~31% 10-year risk is unstudied here.
- Dose: 50-500 mg/day, majority 75-100 mg. A sensitivity analysis restricted to <=100 mg/day gave materially the same result, including the bleeding excess (composite HR 0.89; major bleeding HR 1.54) — so this is a low-dose finding, not an artefact of high doses.
- Age / population: median age 62 (range 53-74); 47% men; 19% had diabetes. Median follow-up 5.0 years — the balance beyond ~5-7 years is not directly characterised.
- Contemporary subset: in trials published since 2000, the CV composite reduction and all bleeding increases persisted, but aspirin was «no longer associated with reduced myocardial infarctions» (Zheng & Roddick, 2019) — consistent with weaker benefit in the modern era of statins and blood-pressure control.
- Cancer: exploratory outcomes were neutral — incident cancer HR 1.01 (0.93-1.08), cancer mortality HR 1.03 (0.96-1.11); no benefit or harm signal, so cancer does not rebalance the decision.
- Certainty: 9 of 13 trials at low risk of bias, heterogeneity low (I2 0-2% for the primary
outcomes), no publication-bias signal (Egger P = .57). The source is gold-tier and internally strong.
confidence: mediumreflects the addition of the USPSTF 2022 what-to-do layer below reaching the same net-balance conclusion — but note the two are not independent (USPSTF’s evidence base is a separately-published systematic review over largely the same trials), so this is authoritative concurrence, not a second independent route; the effect magnitudes above remain Zheng’s.
The guidance layer — USPSTF 2022 and the reversal of routine aspirin
The evidence above is the what-is-true layer; USPSTF 2022 is the authoritative what-to-do layer built on the same class of evidence. It is a recommendation summary (high-tier, borrowed evidence: its effect estimates come from a separately-published systematic review, not restated here), so it is cited for the recommendation grades, age bands, and net-benefit judgement — never for an effect magnitude, which stays with the Zheng meta-analysis above.
The two current recommendations (US Preventive Services Task Force, 2022):
- Adults 40-59 with >=10% 10-year CVD risk — an individual decision (Grade C). «The decision to initiate low-dose aspirin use for the primary prevention of CVD in adults aged 40 to 59 years who have a 10% or greater 10-year CVD risk should be an individual one. Evidence indicates that the net benefit of aspirin use in this group is small.» USPSTF concludes «with moderate certainty» that this group «has a small net benefit».
- Adults 60 or older — recommend AGAINST initiating (Grade D). USPSTF concludes «with moderate certainty that initiating aspirin use for the primary prevention of CVD events in adults 60 years or older has no net benefit».
The recommendation applies to adults >=40 without CVD who are not at increased bleeding risk (no GI-ulcer history, recent bleeding, or bleeding-raising medications). Risk is estimated with the ACC/AHA Pooled Cohort Equations — the only US-validated tool, but one USPSTF flags as imprecise and prone to overprediction, so the 10% threshold is «a starting point to discuss», not a trigger (US Preventive Services Task Force, 2022). Benefit is larger at >15%-20% 10-year risk. Pragmatic dose is 81 mg/d (studied range 50-500 mg/d) (US Preventive Services Task Force, 2022). For those who do start, modeling suggests «data suggest that clinicians and patients should consider stopping aspirin use around age 75 years» (US Preventive Services Task Force, 2022).
Grid mapping — the worked USPSTF cell
This recommendation is the worked audit of Net Benefit and the USPSTF Recommendation Grid against a real topic (the grid page held this open). Both grades read off the moderate-certainty row: Grade C = moderate certainty x small net-benefit magnitude (40-59 at >=10%); Grade D = moderate certainty x zero/negative magnitude (>=60). The C is not thin evidence — it is an affirmatively small net benefit known with moderate certainty, whose action content is shared decision-making; the D is a confident no-net-benefit, not an I (insufficient) (inferred from US Preventive Services Task Force, 2022).
The 2016 -> 2022 reversal, and USPSTF’s stated reason
USPSTF’s 2022 statement replaces its 2016 recommendation and moves it in the de-adoption direction (US Preventive Services Task Force, 2022):
| Feature | 2016 | 2022 |
|---|---|---|
| Start age (individual/C band) | 50-59 at >=10% 10-yr risk | 40-59 at >=10% 10-yr risk (start lowered to 40) |
| Adults 60-69 | individual decision | folded into the >=60 D recommendation |
| Adults >=60 (initiation) | 60-69 individual; >=70 insufficient | Grade D — recommend against (new) |
| Posture | initiate (routine, for eligible) | «selectively based on individual decision-making rather than routinely» |
| Colorectal-cancer benefit | included as a co-benefit | withdrawn — «evidence is unclear whether aspirin use reduces the risk of colorectal cancer» |
The single most consequential change is the new Grade D against initiation at >=60: «There is a new recommendation not to initiate aspirin in adults 60 years or older for primary prevention» (US Preventive Services Task Force, 2022). The routine-to-selective shift is the other: «Aspirin should be initiated selectively based on individual decision-making rather than routinely for all persons in the recommended age and CVD risk group» (US Preventive Services Task Force, 2022).
USPSTF states its own reason for the change (counter-passage read end to end): the reversal rests on new trial evidence (chiefly ASPREE in the healthy elderly), updated analyses of the primary-prevention evidence, and longer-term Women’s Health Study follow-up — an updated evidence base, not a change of method (US Preventive Services Task Force, 2022). In the telos’s five-reason schema this is a self-documented divergence-class-2 (different/updated evidence base): the body updated on new evidence and said so, which is a sound process working, not a defect (inferred from US Preventive Services Task Force, 2022) -> Which Objective Moved This Recommendation.
Why age, not baseline risk, is the switch — the bleeding-age interaction
USPSTF’s net-benefit reasoning turns on a mechanism that concurs with Zheng’s route-(a) analysis above (and, being built on largely the same trials, is not an independent backing — no type-E) (US Preventive Services Task Force, 2022):
- The relative CV effect is not modified by baseline CVD risk; the absolute benefit is larger at higher risk.
- The relative bleeding risk does not differ by age either — but the absolute bleeding incidence, and so the magnitude of bleeding harm, rises with age, and more steeply at >=60.
Because CVD risk itself climbs steeply with age, initiating at older age buys a benefit that the age-driven bleeding harm has already overtaken. USPSTF’s commissioned microsimulation makes the sign flip explicit: net life-years/QALYs are positive for initiation at 40-59, run slightly-negative to negative at 60-69, and are negative at essentially all modelled risk levels for 70-79 (US Preventive Services Task Force, 2022). (Illustrative, net life-years per 1000 at 10% 10-year risk: women +11.4 at 40-49 -> -6.5 at 50-59 -> -13.5 at 60-69 -> -16.6 at 70-79; men +36.1 -> +4.2 -> -3.0 -> -6.9.) This is the same closing-of-the-route-(a)-escape-hatch the Zheng analysis found (harm co-scales with the risk that carries the benefit), here resolved onto an age axis to set the grade cutpoint.
What this is and is not
This page now carries both layers: the what-is-true evidence finding (Zheng) and the what-to-do guidance layer (USPSTF), kept distinct — USPSTF is cited for the recommendation grades and net-benefit judgement, the effect magnitudes remain the meta-analysis’s. What the evidence says on its own terms: in primary prevention the CV benefit is real but small and is offset by a matched bleeding harm, with no mortality benefit and no stratum in which the net turns clearly positive — so the default in an unselected low-risk person is that aspirin is not worth initiating, and any use is an individualised bleeding-risk-vs-benefit judgement. The guidance layer converts that into an operational rule: individualise below 60 at >=10% risk; do not initiate at or above 60.
(inferred from Zheng & Roddick, 2019) Selection, dosing, bleeding-risk screening, PPI co-prescription, and the secondary-prevention indication (established CVD, where the benefit is well established and not in question here) are out of scope — prescriber acts and a different decision.
Limitations (author-stated)
(Zheng & Roddick, 2019) Trial-level (not individual-patient) data; the diabetes subgroup had poorly reported events; endpoint definitions varied across a long time span; 8 trials began randomising over 20 years ago, so older trials may not reflect current background prevention. Hemorrhagic stroke could count in both the CV and bleeding columns.