How should clinicians use the results of randomized trials?
ACP J Club. 1995 Jan-Feb;122:A12. doi:10.7326/ACPJC-1995-122-1-A12
Related Content in the Archives
• Editorial: Applying the findings of clinical trials to individual patients
In a recent editorial in ACP Journal Club, Laupacis and colleagues (1) suggested ways that investigators could present their results to clinicians. Two common approaches are the relative risk (RR) (the ratio of the risk for an adverse outcome among those receiving the treatment of interest compared with that among those not receiving it) and the relative risk reduction (RRR) (the extent to which treatment reduces the risk in comparison with patients not receiving the treatment of interest, or 1 - RR). The absolute risk reduction (ARR), that is, the absolute difference in rates of adverse events in treated and comparison patients, is an important alternative. The number needed to treat (NNT) is an appealing variation of the ARR that indicates the number of patients one needs to treat to prevent 1 adverse outcome, calculated as 1/ARR. For clinical applications, both of these are preferable to the odds ratio, especially when the incidence of adverse outcomes is high (2).
In a study in which the outcome is an adverse event (such as death or myocardial infarction) that a patient either has or does not have, the key result is the extent of risk reduction caused by the treatment, the RRR. The RRR by itself, however, may be misleading. For example, although cholesterol reduction may diminish the relative risk for death from coronary artery disease by the same amount in both asymptomatic men and men who have had myocardial infarction (about 25%), the effect of treating low-risk asymptomatic patients will differ from the effect of treating patients at a much higher risk. Note that we have assumed that the RRR is constant in these 2 groups of patients who have different levels of risk for death. This assumption seems warranted in many instances (3), although it has not been systematically evaluated. For the rest of this discussion, we will continue to assume a constant RR across subgroups that have differing baseline risks.
Why are we so concerned about how clinicians interpret the findings presented by investigators? Because the same results when presented in different ways may lead to different treatment decisions (4-9). For example, Forrow and colleagues (4) showed that clinicians were less inclined to treat patients when trial results were presented as the absolute change in the outcome (ARR) compared with the relative change in the outcome (RR). In 2 similar studies, Naylor and colleagues (5) and Bucher and colleagues (6) found that clinicians rated the effectiveness of an intervention lower when events were presented in absolute terms than when the RRR was used. Moreover, the effectiveness was rated lower when results were expressed as NNT than when the same data were presented as RRR or ARR.
Patients are as susceptible as clinicians to the mode of risk communication (10-12). Investigators found that when patients were presented with a hypothetical life-threatening illness, they were more likely to choose a treatment described in terms of RRR than the equivalent ARR (10). In a similar study, 88% of patients advised of a hypothetical RRR consented to therapy, whereas all other formats (the equivalent ARR, the NNT, and a measure of disease-free survival) elicited significantly fewer endorsements of therapy (42%, 31%, and 40% of patients, respectively) (12).
So, because the manner of presentation affects decisions, is there a correct or an optimal way to assess and apply the results of clinical trials? We suggest that clinicians take 3 steps to apply the results of an investigation to solve a patient problem (13). First, they should consider whether the results are valid (i.e., credible or believable). For articles about therapy, the 3 key validity issues are whether the patients were randomly assigned to different treatments, whether they were analyzed in the groups to which they were assigned, and the extent of follow-up.
Second, when the validity has been ensured, understanding the results of a randomized clinical trial requires that the measures of the effect of the intervention be examined using the RRR, ARR, and NNT. These figures may be in the report itself (we attempt to include them in all abstracts about treatments for ACP Journal Club).
The third step, the application to an individual patient, can only be taken by the attending clinician and requires knowledge of both the trial and the patient. This involves a consideration of both the extent to which the patient resembles those who were enrolled in the study and the patient's risk for the adverse event that the treatment may prevent.
To show all 3 steps in this process, we will now consider 2 patients presenting with acute myocardial infarction and the decision whether to administer tissue plasminogen activator (tPA) or streptokinase. The patients presenting with acute myocardial infarction are a 44-year-old man with no sign of heart failure and a 72-year-old man with heart failure. The study we will use to form our therapeutic decision is the GUSTO trial (14), with a focus on the mortality rate in the 30 days after admission to the hospital of approximately 20 000 patients who received streptokinase and approximately 10 000 who received tPA (11).
In addressing the validity of this trial, we find that patients were assigned to receive the treatment and were analyzed in the groups to which they were assigned; the loss to follow-up was just 0.2% (15). We can conclude, therefore, that the trial results probably represent an unbiased estimate of the true treatment effect.
Turning to the results, the risk for death in the patients who received tPA was 6.3%, and the risk for those who received streptokinase was 7.3%. The RR for death with tPA is, therefore, 6.3/7.3, or 86%, and the reduction in risk (the RRR) is 1 - 0.86, or 14%. For every 1000 patients who received tPA compared with those who received streptokinase, however, 2 more patients had a hemorrhagic stroke. (A subgroup analysis suggested that the treatment effect was larger in those younger than age 75 years [20% RRR] than in those older than age 75 years [6.3% RRR]. We chose to use the total population as the basis for our calculations.)
Now that we understand the results, let us examine the application to our patients. First, the GUSTO trial enrolled the total spectrum of patients across a wide range of ages who presented with myocardial infarction. The results, therefore, are probably applicable to both the 44-year-old man and the 72-year-old man. Next, we must estimate our patients' risks for death in the 30 days after presentation. Studies of the prognosis of patients after myocardial infarction suggest that the risk for death for the 44-year-old man if he receives streptokinase is approximately 2% (16, 17). Administering tPA will reduce this risk by 14% to approximately 1.7%. The ARR for similar patients is 0.02 - 0.017, or 0.003 (0.3%), and the NNT is 1/0.003, or 333. If we treated 1000 such patients with tPA, we would save 3 lives and cause 2 hemorrhagic strokes. Given the additional expense of tPA and the low value most patients would place on life after a disabling stroke, one may well conclude that the right choice is to give streptokinase.
The same prognostic studies suggest that the risk for death in the next 30 days for the 72-year-old man with heart failure is approximately 50%, or 0.50. Administering tPA will reduce the risk by 14% to 0.43, for an ARR of 0.07 and an NNT of 14. If we treated 1000 such patients, we would save 71 lives and cause 2 hemorrhagic strokes. We might well consider that it is better to administer tPA to this second patient.
This example shows that to make the best treatment decision when measures of efficacy are considered, the clinician needs to determine the RRR and ARR from the results of the study (18) and accurately estimate the individual patient's baseline risk, or the risk without treatment. Using these 2 pieces of information, the clinician can calculate the ARR and the NNT that are relevant for that patient. Neither measure is sufficient alone: The RRR by itself can be misleading, and the ARR from a particular study tells us about the average effect of the treatment over all patients in that study, whereas our patient's risk may differ from that average risk.
Although ACP Journal Club will continue to present the RRR and the NNT associated with the ARR in the study (i.e., the NNT for the “average patient”), we encourage investigators to report the baseline risk for patients in the trial and readers to adjust that risk for the patient before them when making their treatment decisions.
Gordon H. Guyatt, MD
Deborah J. Cook, MD
Roman Jaeschke, MD
1. Laupacis A, Naylor CD, Sackett DL.How should the results of clinical trials be presented to clinicians? ACP J Club. 1992 May-Jun;116:A12-14.