When a major clinical trial reports that a new drug cuts the risk of heart attack, stroke, or kidney failure by 20 or 30 percent, the number sounds transformative. Patients hear it and imagine their personal danger has been slashed by a third. Doctors read it and reach for the prescription pad. But a new viewpoint published in eClinicalMedicine by Hussein Zaitoon, Muhammad Abdul-Ghani, and Ralph A. DeFronzo argues that this familiar way of describing treatment effects is quietly distorting medicine, and that cardiorenal trials in particular should be required to report a second, humbler pair of statistics alongside every hazard ratio: absolute risk reduction and the number needed to treat.

The technical distinction at the heart of the argument is deceptively simple. Relative risk reduction describes the proportional decrease in event risk between two groups, calculated as one minus the risk ratio, expressed as a percentage. Absolute risk reduction, by contrast, is the raw difference in cumulative event rates between the control and treatment groups at a specified time point. The number needed to treat is simply the inverse of that absolute reduction: it tells you how many patients must be treated, over that same period, to prevent one additional event. For time-to-event outcomes, the authors stress that these absolute measures should be estimated from cumulative event probabilities using appropriate survival methods, not by crudely subtracting annualized incidence rates. A hazard ratio of 0.79, for example, indicates an estimated 21 percent lower hazard assuming the proportional hazards interpretation holds, and is emphatically not the same thing as a 21 percent reduction in cumulative risk.

The consequences of conflating these measures are vividly illustrated by the PIONEER-6 trial of oral semaglutide. A physician reviewing the study sees a hazard ratio of 0.79 for major adverse cardiovascular events and may reasonably conclude that semaglutide offers a substantial cardiovascular benefit. Yet the trial actually reported such events in 3.8 percent of patients on the drug versus 4.8 percent on placebo over a median follow-up of 15.9 months, an absolute difference of roughly one percentage point and a number needed to treat of approximately 100 over that period. The distinction is not a statistical nicety. It directly shapes how clinicians, patients, guideline committees, and policymakers weigh therapeutic benefit, and it can mean the difference between an enthusiastic endorsement and a measured one.

The problem extends far beyond cardiology. In oncology, the CLEOPATRA trial showed that adding pertuzumab to trastuzumab and docetaxel improved overall survival in HER2-positive metastatic breast cancer with a hazard ratio of 0.68; in an eight-year landmark analysis, overall survival was 37 percent versus 23 percent, an absolute gap of roughly 14 percentage points. In neurology, the AFFIRM trial of natalizumab in relapsing multiple sclerosis reported a 68 percent reduction in annualized relapse rate, while the proportion of patients remaining relapse-free at two years was 67 percent versus 41 percent, an absolute difference of 26 percentage points. Across therapeutic areas, relative and absolute measures tell complementary but strikingly different stories, and the discrepancy is pervasive.

The authors assembled a comparison of seven major cardiorenal outcome trials to show how widely absolute benefit varies even when relative effects look similar. In CREDENCE, canagliflozin produced a hazard ratio of 0.80 for major adverse cardiovascular events, with events in 9.9 percent of treated patients versus 12.2 percent on placebo over a mean follow-up of 2.62 years, a crude absolute difference of 2.3 percentage points; the trial’s own survival-based estimate at 2.5 years was an absolute risk reduction of 2.5 percentage points, corresponding to a number needed to treat of 40. In HARMONY Outcomes, albiglutide yielded a hazard ratio of 0.78, with events in 7.1 percent versus 9.0 percent over 1.6 years, an absolute difference of 1.9 percentage points and a number needed to treat of roughly 53. Viewed through hazard ratios alone, these agents appear broadly comparable; the absolute measures reveal differences in benefit that are strongly driven by baseline event risk and follow-up duration rather than by any intrinsic superiority of one molecule over another.

The kidney field illustrates a further layer of complexity. Canagliflozin in CREDENCE and empagliflozin in EMPA-KIDNEY showed broadly similar relative effects on renal outcomes, with hazard ratios of 0.70 and 0.72 respectively, yet the composite endpoints differed importantly between the trials. CREDENCE used a composite of doubling of serum creatinine, end-stage kidney disease, and renal or cardiovascular death, whereas EMPA-KIDNEY included kidney disease progression defined by a sustained 40 percent or greater decline in estimated glomerular filtration rate, end-stage kidney disease, sustained filtration below 10 milliliters per minute per 1.73 square meters, or renal death, together with cardiovascular death. DAPA-CKD used yet another composite built around a 50 percent or greater sustained decline in filtration rate. These differing definitions, eligibility criteria, baseline risk profiles, and follow-up durations all influence absolute event rates, and EMPA-KIDNEY enrolled a broader and overall lower-risk population. Apparent differences in absolute benefit across these trials therefore cannot be read as evidence of differential drug efficacy.

The stakes are rising because the volume of cardiovascular and kidney outcome trials involving GLP-1 receptor agonists and SGLT2 inhibitors has expanded rapidly, and these therapies are increasingly prescribed across cardiology, nephrology, endocrinology, and primary care. As treatment indications broaden, clinicians need measures that communicate not merely whether a therapy works but how much benefit an individual patient can realistically expect. Absolute measures are often already partially available in journals that report event rates alongside hazard ratios, from which attentive readers can estimate absolute effects themselves, but explicit reporting would make the calculations immediately accessible and reduce the risk of misinterpretation. Methodological evidence indicates that numbers needed to treat derived from differences in incidence rates can be biased, whereas estimates built on cumulative event probabilities from survival analysis are more appropriate.

One domain where absolute effects prove especially valuable is in weighing efficacy against safety. Relative risks alone are insufficient for balancing events caused by treatment against events prevented. The ASCEND trial, which tested aspirin against placebo in patients with type 2 diabetes, is an instructive example: the absolute reduction in serious vascular events was 1.1 percentage points, while the absolute increase in major bleeding was 0.9 percentage points. Placed side by side, those figures make the delicate balance between benefit and harm immediately apparent in a way that any pair of relative percentages never could. The recently published SOUL trial of oral semaglutide reinforces the point about method as well: over a median follow-up of 49.5 months, major adverse cardiovascular events occurred in 12.0 percent of treated patients versus 13.8 percent on placebo, a crude absolute difference of 1.8 percentage points, while the prespecified survival-based estimate at three years was an absolute risk reduction of 2.0 percentage points and a number needed to treat of 50. The corresponding annualized incidence rates, 3.1 and 3.7 events per 100 person-years, form an incidence rate difference that should never be converted directly into a conventional number needed to treat by taking its reciprocal.

The authors are careful to acknowledge that their argument cuts both ways. Relative measures have genuine and distinct advantages: where there is no strong evidence of effect modification, relative risk reductions are generally more generalizable beyond the trial population, because absolute event rates vary substantially with baseline risk, secular trends, geography, and selection criteria. For an individual patient, the most appropriate estimate of absolute benefit comes from applying an applicable relative treatment effect to that patient’s own baseline risk over a specified time horizon, rather than simply importing the trial’s figure. Moderate to large relative reductions in the 20 to 39 percent range represent substantial treatment effects even in lower-risk populations and can have major consequences for population-level disease burden. The proposal, then, is not that absolute measures should supplant relative ones, but that they should travel together.

There are also inherent limits to the number needed to treat itself. It is inherently time-specific, so a figure calculated over one year is not equivalent to one calculated over five, because treatment effects and baseline event rates change over time. Cardiorenal trials typically run for two to five years, yet in practice these therapies are used far longer, and longer treatment may yield greater cumulative benefit, though this cannot be assumed to accrue linearly given changing hazards, competing risks, adherence, and discontinuation. Trial populations are highly selected and may differ substantially from routine clinical practice. The proposed solution is straightforward: journals should routinely require absolute risk reduction, number needed to treat, and baseline event rates alongside relative measures, with the time horizon and estimation method clearly stated, ideally at standardized time points such as one, three, and five years. Guidelines and regulators should follow suit. Transparent communication of treatment effects, the authors conclude, is essential if clinicians and patients are to make genuinely informed decisions.

Subject of Research: Reporting of absolute risk reduction and number needed to treat in cardiorenal outcome trials

Article Title: Relative risk reduction is not enough: why cardiorenal trials should routinely report absolute risk reduction and number needed to treat

Article References: Zaitoon, H., Abdul-Ghani, M., & DeFronzo, R. A. (2026). Relative risk reduction is not enough: why cardiorenal trials should routinely report absolute risk reduction and number needed to treat. eClinicalMedicine, 101, Article 104258. https://doi.org/10.1016/j.eclinm.2026.104258

Image Credits: AI Generated

DOI: 10.1016/j.eclinm.2026.104258

Keywords: absolute risk reduction, number needed to treat, hazard ratio, cardiorenal outcome trials, SGLT2 inhibitors, GLP-1 receptor agonists, semaglutide, chronic kidney disease, major adverse cardiovascular events, clinical trial reporting, eClinicalMedicine, evidence-based medicine