Clinically useful measures of the effects of treatment.

In the abstracts in this journal that describe effective treatments, we provide our readers with numbers that summarise their clinical effect. These include the number of patients you need to treat to prevent I adverse outcome (NNT) and both die absolute risk reduction (ARR) and relative risk reduction (RRR) in the occurrence of adverse outcomes achieved by active therapy. In this EBM Note, we explain these numbers for our readers and use them to begin a glossary of terms that will appear in each issue. In the first issue of Evidence-Based Medicine, we presented the results of die Diabetes Control and Complications Trial (DCCT) (1) into the effect of intensive diabetes therapy on the development and progression of neuropathy. In that trial, confirmed neuropathy developed among 9,6% of patients randomly assigned to usual care (1 or 2 insulin injections/d to prevent gylcaemic symptoms; we will call this rate "C" for "control") and among 2.8% of patients randomly assigned to intensive therapy (insulin pump or > 3 injections/d; we call this rate "E" for experimental). This difference was statistically highly significant, but how might this treatment effect be expressed in terms of its clinical significance? The traditional measure of this effect is the proportionai or "relative" risk reduction (abbreviated RRR in our journal), calculated as (C E)/C. In this example, the RRR is (9.6% 2.8%)/9.6% or 71%; intensive therapy reduced the risk for developing neuropathy by 71%. Why not confine our description of the clinical significance of this result to die RRR? The reason is that the RRR fails to discriminate huge absolute treatment effects (10 times those observed in this trial) from those that are trivia! (1/10 000 of those observed here). For example, if the rates of neuropathy were 10 times those observed in this trial, and a whopping 96% of control patients and 28% of intensively treated patients developed neuropathy, the RRR would remain unchanged: RRR = (96% 28%)/96% or 71 %. And if a trivial 0.00096% of control and 0.00028% of intensively treated patients developed neuropathy, the relative risk reduction is as before: RRR still = (0.00096% 0.00028%)/0.00096 = 71%! This is because the RRR discards the underlying susceptibility (or "baseline risk") of patients entering randomised trials; as a result, the RRR cannot discriminate huge risks and benefits from small ones. In contrast to these nortdiscriminating RRRs, the absolute differences in the rates of neuropathy between control and experimental patients (C E) clearly do discriminate between these extremes, and this measure is called the absolute risk reduction or ARR. In the DCCT, the ARR or (C E) = 9.6% 2.8% = 6.8%; in the extremely high hypothetical example, in which 96% of control patients and 28% of intensively treated patients developed neuropathy, the ARR or (C E) = 96% 28% = 68%; in the extremely low hypothetical example, in which a trivial 0.00096% of control and 0.00028% of intensively treated patients developed neuropathy, the ARR or (C E) = 0.00096% 0.00028% = 0.00068%. These ARRs retain the underlying susceptibility of patients and provide more detailed information than RRRs. But, unlike RRRs that can be recalled as whole numbers, ARRs are decimals and are therefore difficult to remember and do not slip easily off the tongue at the bedside. If, however, we divide the ARR into 1 (i.e., if we "invert" the ARR or "take its reciprocal" so that it becomes 1/ARR), we generate a very useful number because it represents the number of patients we need to treat (NNT) with the experimental therapy in order to prevent 1 bad outcome. In die DCCT, we would generate die number of persons widi diabetes we would need to treat widi the intensive regimen in order to prevent 1 from developing neuropathy. In the trial, the N N T is 1/ARR or 1/6.8% or 14.7; we usually round tiiat number upward (in this case, to 15), and we now can say that for every 15 patients who are treated widi the more intensive insulin regimen, 1 is prevented from developing diabetic neuropadiy. Is 15 a large or a small number of patients that need to be treated to prevent 1 bad outcome? As with many important matters in medicine, the answer has to do widi clinical significance, not statistical significance. This N N T of 15 certainly is for smaller than the number of patients we would need to treat in die extremely low hypothetical example, in which 1/ARR becomes 1/ 0.00068%, or an. N N T of more than 147 000, a figure so vast that we cannot imagine anyone judging it to be worth die effort. We can get a better idea by comparing this N N T of 1S with that for other interventions we are familiar with in medicine. In doing so, we add the dimension of the duration of dierapy: in the DCCT, treatment continued for an average of 6.5 years, meaning that we need to treat about 15 persons with diabetes for about 6.5 years with an intensive insulin regimen to prevent 1 from developing neuropathy. How does this compare with other treatments, over other durations, for other conditions? Beginning on an optimistic note, only about 20 patients with chest pain who appear to be having heart attacks need to be treated widi streptokinase and aspirin to save a life at 5 weeks. On die other hand, about 70 elderly persons with hypertension need to