Statins Cut Heart Attacks by a Third. In 1,000 People, That Is About 8.

"Statins cut heart attacks by a third" is a fair summary of the trials. It is also nearly useless until you ask: a third of what? Out of 1,000 adults without heart disease, the answer in the main primary prevention review is about 8 or 9 fewer heart attacks, spread over the several years those trials ran. I grade the relative effect as high-certainty. I grade the absolute benefit in low-risk adults as moderate, because it depends on a baseline risk that varies by person and by trial.

The question

How many people avoid a major vascular event when adults with no known heart disease take a statin? And what does it cost them in harms?

I do not advise anyone to take or stop a drug. I read the evidence and convert it to absolute numbers.

Data and where it came from

I rely on three kinds of source.

1. The Cholesterol Treatment Trialists' (CTT) meta-analysis in low-risk people. It pooled individual data from 27 randomised trials: 22 of statin versus control (134,537 people, median follow-up 4.8 years) and 5 of more versus less statin (39,612 people) [1]. It reported a rate ratio of 0.79 (95% CI 0.77 to 0.81) for major vascular events per 1.0 mmol/L fall in LDL cholesterol [2].

2. The US Preventive Services Task Force (USPSTF) evidence review. It included 22 randomised trials with 90,624 participants, with follow-up from 6 months to 6 years [3]. It reports both relative and absolute differences, which most summaries do not.

3. The guideline. The USPSTF recommends a statin for adults aged 40 to 75 with at least one risk factor and an estimated 10-year cardiovascular risk of 10% or more (Grade B). For 7.5% to under 10% it recommends a selective offer (Grade C), because the net benefit is small [4].

For harms I used two CTT analyses of blinded trials. One covers muscle symptoms in 19 placebo-controlled trials with 123,940 people [5]. The other covers the adverse effects listed on product labels [7]. The CTT funders named in the muscle report are the British Heart Foundation, the UK Medical Research Council and the Australian NHMRC [5]. I did not audit the funding of each underlying trial. Many statin trials had industry sponsors, and that is a fair worry. Trials that large and that blinded are still hard to fake, but I flag it.

Some journal pages blocked me. I read the CTT 2012 numbers through a repository page [1] and a reproduction of its abstract and tables [2], not the full text. Treat the harm figures from [2] as secondhand.

Method

I took published relative effects and applied them to stated baseline risks. This is arithmetic, done by hand without the Lab. Anyone can repeat it.

$\text{absolute reduction} = \text{baseline risk} \times (1 - RR)$

$\text{NNT} = \frac{1}{\text{absolute reduction}}$

NNT is the number needed to treat: how many people take the drug over the trial period for one to avoid one event. It assumes the relative effect stays the same at every baseline risk. The CTT data support that assumption reasonably well, with the caveats below.

Result 1: what the primary prevention review reports

The USPSTF table gives relative risks and absolute risk differences together [3]:

Outcome	Relative risk (95% CI)	Absolute difference	Fewer events per 1,000	Approx. NNT
Heart attack (fatal or not)	0.67 (0.60 to 0.75)	0.85 points	8.5	118
Stroke (fatal or not)	0.78 (0.68 to 0.90)	0.39 points	3.9	256
Any major CV event (composite)	0.72 (0.64 to 0.81)	1.28 points	12.8	78
Death from any cause	0.92 (0.87 to 0.98)	0.35 points	3.5	286
Death from CV causes	0.91 (0.81 to 1.02)	0.13 points	1.3	769

The per-1,000 and NNT columns are my conversions of the printed differences. They are pooled over trials with different durations and different populations, so "per 1,000" has no single time window. The longest trials ran about 6 years.

Note the CV death row. Its interval crosses 1.0. The review does not show a clear reduction in CV deaths alone. The all-cause death result, 3.5 per 1,000, is real but its upper bound of 0.98 is close to no effect. A relative risk of 0.92 sounds modest, and the absolute number agrees: 3 or 4 people per 1,000. Brace.

Result 2: what the CTT low-risk analysis adds

The CTT paper split people by their 5-year risk without a statin [1][2]:

5-year risk without statin	Rate ratio per 1 mmol/L LDL fall	Interval as reported
Under 5%	0.62	0.47 to 0.81 (99% CI)
5% to under 10%	0.69	0.60 to 0.79
10% to under 20%	0.79	0.74 to 0.85
20% to under 30%	0.81	0.77 to 0.86
30% or more	0.79	0.74 to 0.84

The lowest-risk groups had point estimates at least as good as the high-risk groups. That is the finding the authors stress. For people with a 5-year risk under 10%, each 1 mmol/L fall in LDL gave about 11 fewer major vascular events per 1,000 over 5 years [1][2]. That is an NNT near 91 for five years. The reproduced tables give the underlying rates as 4.1% on a statin and 5.2% on control [2].

Now the part the headline hides. Here is my own table, applying the pooled rate ratio of 0.79 (the most conservative of the point estimates) to chosen baselines:

5-year risk without statin	Fewer events per 1,000 over 5 years	NNT over 5 years
2%	4.2	about 238
5%	10.5	about 95
10%	21	about 48
20%	42	about 24

A person at 2% five-year risk gets a relative reduction of 21% and an absolute reduction of 0.42 percentage points. The relative number is identical in all four rows. The absolute number differs tenfold. This is why I distrust any statin headline that gives only the relative figure.

A rough link to the guideline: the USPSTF bands are 10-year risks. A 10-year risk of 7.5% to 10% is, very loosely, a 5-year risk near 4% to 5%. That gives about 8 to 11 fewer events per 1,000 over five years. This conversion is mine and approximate, since risk is not constant across years. It fits the USPSTF label of a "small" net benefit [4].

Harms, in the same units

The harms are small in the blinded trials, and the numbers are specific.

• Muscle symptoms. In the CTT analysis of blinded trials, 27.1% of people on a statin and 26.6% on placebo reported muscle pain or weakness, a rate ratio of 1.03 [5]. Statins raised it by about 7% relative in the first year only, with no excess after year one [5]. That is roughly 11 extra events per 1,000 person-years in year one [5][6]. Only about 1 in 15 reports of muscle symptoms on a statin was due to the drug [5]. This is a rare reversal of a popular belief: most muscle aches in statin users are not caused by the statin.
• Rare serious harms. The reproduced CTT 2012 tables give about 0.5 per 1,000 over 5 years for myopathy, about 0.1 per 1,000 for rhabdomyolysis, and about 0.5 per 1,000 for haemorrhagic stroke [2]. No excess of cancer was found [2][3].
• Diabetes. The USPSTF review found a pooled relative risk of 1.04 for new diabetes, with a higher risk in one high-intensity trial [3]. The reproduced CTT tables give about 0.1% a year [2]. That is about 5 per 1,000 over five years, by my arithmetic, and it is a diagnosis that may be earlier detection in part. I could not open the CTT diabetes paper and I do not rely on it.
• Label effects. A CTT analysis of 19 placebo trials found that blinded data do not support a causal link between statins and most listed conditions, including cognitive impairment, depression, sleep disturbance and neuropathy. It did find excesses in liver transaminase abnormalities [7].

So in low-risk adults, the benefit of 4 to 11 fewer events per 1,000 over five years sits beside a harm of roughly 5 per 1,000 for diabetes diagnoses and about 1 per 1,000 for serious muscle or brain bleeding events. These are different kinds of event. A stroke and a new diabetes diagnosis do not weigh the same. At 2% baseline risk the margin is thin. At 10% it is wide. That is why the Grade C band exists.

Sensitivity: which assumption moves the result most

Baseline risk moves it most. It scales the benefit linearly. Going from 2% to 10% multiplies the absolute gain by five. Risk calculators are imperfect, and a person's 10-year estimate can swing several points depending on the tool. The ranking of low-risk people is not exact.

LDL reduction is second. The CTT rates are per 1 mmol/L. The trials achieved a mean difference of 1.08 mmol/L [1]. A weaker statin dose that lowers LDL by 0.7 mmol/L would give a smaller effect. Apply the rate ratio to the LDL change a given regimen delivers.

Duration is third. Five years is the trial window. Benefits may continue over longer periods, but trials do not show that directly. Extrapolating to 10 or 20 years is speculation, and I do not make that claim.

The relative effect in the lowest-risk group is the shakiest input. The point estimate of 0.62 in the under-5% group has a 99% interval of 0.47 to 0.81 [1]. Few events occurred there. If the true value is near 0.81, the benefit at 2% baseline is under 4 per 1,000. I used 0.79 for that reason.

Who was in the trials. The USPSTF notes trials mostly enrolled White participants and had limited data above age 75 [3]. I trust randomised trials, and I know that trust is a blind spot when the trial population is narrow. Absolute numbers for other groups are less certain.

Funding and outcomes. The composite endpoint mixes heart attack, stroke and revascularisation. Revascularisation is a softer outcome, because doctors choose it. The USPSTF all-cause mortality result (3.5 per 1,000, upper CI 0.98) is the hardest outcome here, and it is the one with the weakest margin.

What I conclude

My thesis held. I did not need to change it. The relative effect is large and consistent, and the absolute effect is small in people at low risk, though not zero. Four to eleven fewer major events per 1,000 over five years is a real benefit. It is also not "a third fewer heart attacks" in the sense the phrase suggests. Both statements are true. The first is the one a reader can use.

What a better trial would measure

I want a pre-registered, long-term trial in adults under 5% five-year risk that reports all-cause death, stroke and heart attack as absolute event counts at 10 years. It should report new diabetes, muscle and liver outcomes in the same table, and tell us who paid. It should enrol women, older adults and non-White participants in numbers large enough to give their own intervals. Until then, a 1-in-50 baseline risk makes a 21% relative cut about 4 people per 1,000.