Background

Randomized clinical trials are the gold standard for informing the effectiveness of therapeutic interventions, and often report a variety of effect measures that readers should consider [1]. Commonly used measures for dichotomous outcomes in randomized controlled trials include absolute risk, risk difference, relative risk, relative risk reduction, and odds ratio [1, 2]. These measures can be broadly categorized as conveying either absolute or relative effects and, depending on the choice of measure(s) used by authors, treatment effects may appear larger or smaller despite being based on the same data [1, 2]. This may lead clinicians to make different treatment decisions depending on how the results are presented [2,3,4]. Thus, it is imperative that clinicians understand how to interpret trial data, independently of the narrative of the sponsoring group, and be able to convey this information to patients to optimize shared decision-making.

Throughout this editorial we demonstrate how to identify, calculate and interpret different outcome measures for a two-arm, parallel-group, superiority randomised controlled trial described in Table 1 ref. [5]. We acknowledge that other trial designs such as non-inferiority trials may require alternative approaches to interpreting results. We describe and compare the measures of effect of laser trabeculoplasty compared to 0.5% timolol eye drops for glaucoma [5]. The dichotomous outcome, treatment failure, was assessed at 12 months post-randomization and defined as an intraocular pressure (IOP) greater than the target pressure depending on glaucoma severity [5]. We used a 2 × 2 table (Table 2) to capture information on treatment failure to aid in the visualization of study findings and the calculation of the different measures of effect using the equations presented in Box 1. Findings from the trial are presented in Table 3.

Table 1 Study characteristicsa.
Full size table

Measure of absolute effect

Absolute risk

The absolute risk is defined as the probability that an event will occur [1, 2, 6]. We commonly refer to the risk of the adverse outcome in the control group as the baseline risk. Based on our example trial, the risk of treatment failure among participants who received timolol eye drops was 69% (121/176 = 0.69) compared to 39% (64/163 = 0.39) in participants who underwent laser trabeculoplasty. In other words, clinicians can expect that 69 individuals out of every 100 using timolol eye drops will experience treatment failure compared to 39 individuals out of every 100 who undergo laser trabeculoplasty.

Risk difference

The risk difference (RD), also known as the absolute risk reduction (ARR) [1, 2], allows trialists to describe the difference in outcome rates using absolute terms (i.e., absolute risk in control group (timolol) – absolute risk in treatment group (laser)). Accordingly, the RD in our working example is 0.69-0.39 = 0.3. This can be interpreted as a 30% ARR in treatment failure with laser trabeculoplasty compared to timolol.

Measures of relative effect

Relative risk and relative risk reduction

In comparison to the measure of absolute effect, measures of relative effect are expressed as a ratio where the expected outcome in one group is assessed relative to that in the other [1, 2, 7]. The relative risk, also known as the risk ratio (RR), is the ratio of the risk or probability of an event in the intervention group (laser) to the risk of an event in the control group (timolol). As per the equation provided in Box 1 and the calculated risks per group, the RR in this case would be 0.39/0.69 = 0.57 or 57%. This may be interpreted as the risk of treatment failure with laser trabeculoplasty is a little more than half compared to patients treated with timolol eye drops. Another way to present this information is by using the relative risk reduction (RRR), which presents an estimation of how much less the risk is in the intervention group compared to the control group and is calculated as 1-RR. In our example, the RRR = 1-0.57 = 0.43 or 43% meaning that laser trabeculoplasty decreases the risk of treatment failure by a little less than half compared to timolol eye drops.

Table 2 A 2 × 2 table.
Full size table

Odds ratio

Instead of assessing the risk of an event, we could look at the odds of an event occurring with or without the intervention by calculating the odds ratio (OR). The OR is the comparison (or ratio) of the odds of an event in the intervention group (laser) with the odds of that event in a control or reference group (timolol eye drops) [1, 2, 8, 9]. Going back to our example, the odds of treatment failure with laser trabeculoplasty is 64 (treatment failure) divided by 99 (treatment success) or 64/99 = 0.65, and the odds of treatment failure in the timolol eye drop group are 121/55 = 2.2. This yields an OR for the comparison of laser versus eye drops of 0.65/2.2 = 0.3. An OR < 1 indicates that there is a decrease in the odds of treatment failure occurring in those undergoing laser therapy, which indicates benefit of the intervention.

Clinicians generally find it is easier to interpret RRs that present event probabilities compared to ORs that describe the odds of events [1, 7, 9]. Fortunately, the OR and RR are interchangeable when the baseline risk is very low (i.e., the outcome is rare) [1, 7, 9]. Specifically, as the risk falls below 20%, odds and risk are more and more similar and nearly the same when the risk is under 10% ref. [9]. However, with higher risks substituting one for the other could be misleading and may over or under-estimate the treatment effect. In cases where the baseline risk is high, it is possible to calculate the risk from the odds using the following equation: risk = odds/ (odds + 1) ref. [10].

Table 3 Results in a 2 × 2 table from a randomized trial of selective laser trabeculoplasty compared with timolol eye drops for controlling intraocular pressure in adults with open angle glaucoma living in Tanzaniaa.
Full size table

Measures to focus on for clinical practice

It is essential to differentiate between absolute (e.g., RD) and relative (e.g., RR or RRR) measures of effect, as relative measures may provide exaggerated effects or misleadingly large estimates especially when the baseline risk is low [1, 2, 6]. Let us assume two different risk groups, based on the severity level of glaucoma, a high-risk group with a hypothetical baseline risk of 20% for treatment failure and the low-risk group with a baseline risk of 2% (Fig. 1). In the high-risk group, a 50% RRR post intervention would decrease the absolute risk from 20% to 10%. A RD of 10% in terms of effect of the intervention is impressive. However, in the low-risk group with a baseline risk of 2%, a 50% RRR would decrease the risk from 2 to 1%, thus giving us a 1% RD, which is considerably less impressive and may not be worthwhile when considering other potentially relevant factors such as patient burden, costs, and potential harms.

Fig. 1: Absolute Risk of Treatment Failure in Two Different Risk Groups Based on Severity Level.
figure 1

Example of a 50% relative risk reduction across different baseline risks comparing control groups (crosshatch) vs treatment groups (solid).

Full size image

Despite the importance of acquiring absolute measures of effect, doing so may require reading more than just the published abstract. We previously reviewed 96 Cochrane and 94 non-Cochrane systematic reviews of randomized trials and found that while 78% reported relative measures of effect for beneficial outcomes, only 23% reported absolute effects for these outcomes. The results for harms were worse; whereas 87% of reviews reported relative measures of effect for harms, only 13% reported absolute effect estimates for harms in their abstract. There was no difference in the proportion reporting absolute effects in the abstract between Cochrane and non-Cochrane reviews [11].

Considering how findings may be interpreted when presented with different effect measures, clinicians should be cautious about relative measures of effect and search out absolute measures. Ideally, trials should present both the RR and the RD so that two pieces of information are provided: the effect of the intervention (RR) and the difference in absolute risk with and without treatment (RD). However, if a trial fails to do so, which is often the case, clinicians can derive the RD by applying relative measures of effect to the patient’s estimated baseline risk to inform patient discussions and guide clinical decision-making.