Clinical article · Specialties, Pathways and documents

Reading a trial in ten minutes: absolute risk, number needed to treat and certainty

Written for
Hospital physicians, emergency doctors and residents
Reading time
7 minutes
Published
Updated

Written by the Sana Health team for clinicians. It summarises published guidance, is not a guideline and has not been peer reviewed. Check current local protocols before acting.

Key points

  • Check first that the trial's population, comparator and primary outcome match the question you are asking.
  • Convert relative effects into absolute risk differences and a number needed to treat, always with the time frame.
  • Apply the relative effect to your own patient's baseline risk; the same relative benefit means very different absolute benefit in low-risk and high-risk patients.
  • Look for the main sources of bias: allocation, blinding, loss to follow-up, selective reporting and early stopping for benefit.
  • Treat subgroup claims with caution unless they were prespecified, plausible and consistent with other evidence.
  • Finish with a one-line bottom line that states the effect, the time frame and how certain the evidence is.

Minute one: what question did the trial answer?

Before the results, read the methods for the population, the intervention, the comparator and the primary outcome. A trial of patients younger, fitter or at higher risk than yours answers a different question, and an active comparator at a low dose flatters the new treatment. CONSORT 2025 sets out what a trial report should contain, which makes it a useful checklist for what to look for and what is missing [1]. Noninferiority trials ask whether a treatment is not unacceptably worse than another and have their own reporting requirements, including the prespecified margin [2].

Minutes two to four: the outcome and the size of the effect

Ask whether the primary outcome matters to patients. Changes in blood pressure, HbA1c or tumour size are surrogates, and treatments that improve a surrogate have not always improved survival or symptoms [3]. Composite outcomes increase statistical power but can be driven by their least important component; in PARADIGM-HF, for example, the primary outcome combined death from cardiovascular causes with hospitalisation for heart failure [4], and each component deserves a look in its own right.

Then convert the result into absolute terms. The figures in the table are illustrative, not taken from any trial.

MeasureHow it is calculatedIllustrative value
Event rate, controlEvents divided by patients in the control group10% over 3 years
Event rate, treatmentEvents divided by patients in the treatment group8% over 3 years
Relative riskTreatment rate divided by control rate0.80
Relative risk reduction1 minus the relative risk20%
Absolute risk reductionControl rate minus treatment rate2 percentage points
Number needed to treat1 divided by the absolute risk reduction50 for 3 years to prevent one event

The relative effect is usually the most transportable part of a trial, but your patient's absolute benefit depends on baseline risk. With the same relative risk of 0.80, a patient whose 3-year risk is 5% gains an absolute reduction of about 1 percentage point, a number needed to treat of about 100. Relative figures make treatments look more impressive than absolute ones, and evidence on communicating risk supports presenting absolute numbers to patients [5]. Calculate the number needed to harm for the main adverse effect in the same way, and set the two side by side.

Minutes five to seven: could bias explain the result?

  • Allocation: was the sequence random and concealed from those enrolling patients?
  • Blinding: were patients, clinicians and outcome assessors blinded? Unblinded studies are particularly prone to overstating subjective outcomes and adverse effects; statin product labels, for example, list effects drawn mainly from non-randomised and non-blinded studies [6].
  • Attrition: how many patients were lost, and was the analysis by intention to treat?
  • Selective reporting: does the primary outcome match the registered protocol?
  • Early stopping: trials stopped early for benefit often report implausibly large effects and give too little information about the decision to stop [7].

Two further checks are quick. The fragility index counts how many patients would need a different outcome for a statistically significant result to lose significance [8], and a small number warns that the finding is fragile. Spin, wording that presents a non-significant primary result favourably, is common in trial reports, so read the primary outcome before the abstract's conclusion [9].

Minute eight: subgroups and applicability

Investigators often claim that effects differ between subgroups when a true difference is unlikely. Credibility criteria, summarised in the ICEMAN instrument, include a prespecified hypothesis with a stated direction, support from earlier evidence, a low probability that chance explains the difference, and consistency across related outcomes [10]. Applicability is a separate question, and it is why individual participant data analyses exist: one pooled six blood pressure trials to quantify how the balance of benefit and harm from intensive control varies with the target and with patient characteristics [11].

Read the confidence interval, not only the p value. If the interval includes both a clinically important benefit and no effect, the trial has not settled the question, whatever the headline says.

Minutes nine and ten: certainty and the bottom line

The GRADE approach rates the certainty of evidence for each outcome as high, moderate, low or very low. Randomised evidence starts high and is rated down for risk of bias, imprecision, inconsistency between studies, indirectness to the question, or suspected publication bias. Guideline panels use the same framework to identify patient-centred outcomes, synthesise the evidence and grade their recommendations [12]. A single trial seldom gives high certainty for every outcome.

  1. State the population and the comparison in one phrase.
  2. Give the absolute effect on the outcome that matters, with its time frame and number needed to treat.
  3. Give the main harm in the same terms.
  4. Rate the certainty and name the main reason for any downgrade.
  5. Say whether it changes practice for your patients, and whether a guideline has already weighed it.

How Sana Health helps

Evidence Review restates a clinical question as population, intervention, comparator and outcomes, gives the position of current guidelines, then sets out the pivotal trials and systematic reviews in a table with design, population, comparator, key result and limitations. It reports absolute effects and numbers needed to treat or harm where the records support them, appraises bias, precision, consistency and directness, and rates the certainty of each main outcome in the manner of GRADE. Ongoing trials from ClinicalTrials.gov are listed, and every reference is read from the source databases.

Questions

How do I calculate the number needed to treat?

Subtract the treatment group's event rate from the control group's to get the absolute risk reduction, then divide 1 by it. A fall from 10% to 8% is a 2 percentage point reduction, so 50 patients need treatment for the trial's duration to prevent one event. Always report the time frame with the number.

Why is the relative risk reduction not enough on its own?

Because absolute benefit depends on baseline risk. A 20% relative reduction prevents one event in 50 patients at 10% risk but one in 500 at 1% risk, and the harms of treatment may be the same in both.

What makes a subgroup result believable?

A subgroup effect is more credible when it was prespecified with a stated direction, is supported by earlier evidence and biological reasoning, is unlikely to be explained by chance, and is consistent across related outcomes and trials. Most post hoc subgroup claims do not meet these criteria.

References

  1. CONSORT 2025 Statement: Updated Guideline for Reporting Randomized Trials (opens in a new tab)JAMA, 2025
  2. Reporting of noninferiority and equivalence randomized trials: extension of the CONSORT 2010 statement (opens in a new tab)JAMA, 2012
  3. Surrogate end points in clinical trials: are we being misled? (opens in a new tab)Ann Intern Med, 1996
  4. Angiotensin-neprilysin inhibition versus enalapril in heart failure (opens in a new tab)N Engl J Med, 2014
  5. Evidence-based risk communication: a systematic review (opens in a new tab)Ann Intern Med, 2014
  6. Assessment of adverse effects attributed to statin therapy in product labels: a meta-analysis of double-blind randomised controlled trials (opens in a new tab)Lancet, 2026
  7. Randomized trials stopped early for benefit: a systematic review (opens in a new tab)JAMA, 2005
  8. The Fragility Index in Multicenter Randomized Controlled Critical Care Trials (opens in a new tab)Crit Care Med, 2016
  9. Spin and fragility in randomised controlled trials in the anaesthesia literature: a systematic review (opens in a new tab)Br J Anaesth, 2023
  10. Evidence-based Urology: Subgroup Analysis in Randomized Controlled Trials (opens in a new tab)Eur Urol Focus, 2021
  11. Benefit-harm trade-offs of intensive blood pressure control versus standard blood pressure control on cardiovascular and renal outcomes: an individual participant data analysis of randomised controlled trials (opens in a new tab)Lancet, 2025
  12. AGA Living Clinical Practice Guideline on Pharmacological Management of Moderate-to-Severe Ulcerative Colitis (opens in a new tab)Gastroenterology, 2024