Application of Number Needed to Treat (NNT) - GOLD

Treat Respir Med 2006; 5 (2): 79-84
LEADING ARTICLE 1176-3450/06/0002-0079/$39.95/0
© 2006 Adis Data Information BV. All rights reserved.
THIS MANUSCRIPT IS PROVIDED IN CONFIDENCE TO DETERMINE REPRINT INTEREST ONLY AND SHOULD NOT BE DISTRIBUTED
EITHER INTERNALLY OR EXTERNALLY VIA PRINT OR ELECTRONIC MEDIA FOR OTHER THAN THE STATED PURPOSE.
Application of Number Needed to Treat (NNT)
as a Measure of Treatment Effect in
Respiratory Medicine
Mario Cazzola
Unità di Pneumologia ed Allergologia e Settore di Farmacologia Clinica Respiratoria, Dipartimento di Pneumologia,
Ospedale ad Alta Specializzazione A. Cardarelli, Naples, Italy
Abstract
Presentation of clinical data can have a profound effect on treatment decisions, and there is a need for
measures that are objective, have clinical relevance, and are easily interpreted. Relative risk is often used to
summarize treatment comparisons, but does not account for variations in baseline risk profiles and does not
convey information on absolute sizes of treatment effects. Absolute risk reduction gives this information, but the
data are dimensionless and abstract, and lack a direct connection with the clinical environment.
The number needed to treat, or NNT, has been developed to address this issue. NNT is the reciprocal of the
absolute risk reduction associated with an intervention, and may also be calculated as 100 divided by the absolute
risk reduction expressed as a percentage. The result is the number of patients who would have to receive
treatment for one of them to benefit or to avoid an adverse outcome over a given period of time. Since its
introduction, the concept of NNT has been expanded to include number needed to harm (NNH), which illustrates
adverse events or other undesirable outcomes associated with treatment, and the epidemiologic tool of number
needed to screen.
NNT has been used to describe treatment effects from many clinical trials. A recent example illustrates
benefit of inhaler therapy combining a long-acting β2-agonist (LABA) and corticosteroid for COPD over
treatment with LABA alone. NNT has also been extended to systematic reviews and meta-analyses, where it has
been used to rank different treatments where baseline profiles, treatment outcomes and time periods under
examination are similar.
NNT is therefore a concise and easily understood tool for quantifying treatment efficacy, particularly when
applying trial results to the clinic setting.
It has been clear, since publications in the early to mid-1990s, treatment decision on the basis of outcomes results reported as
that the results of clinical studies, and the method of reporting trial relative rather than absolute changes. [2]
results, influence clinicians, policy makers and patients. [1-4] Impor- This has led to an increased emphasis on the need for
tantly, perception of the magnitude of any intervention can be evidence-based prescribing. [5] However, despite ongoing improveinfluenced
markedly by the way in which measurements of effect ments in the way information is presented, it is still often difficult
are reported.
for prescribers and decision makers to decide when clinical evidence
As an example, Naylor et al. [4] found that clinicians’ views of
is strong enough to justify a change in practice.
therapeutic effectiveness varied according to the endpoint used. The present review therefore examines this issue, focusing on
For example, ratings were lower when results were reported as one particular endpoint – number needed to treat (NNT) – that has
absolute versus relative differences. [4] In another piece of research been developed in an attempt to provide an easily interpreted and
published at around the same time, 108 of 235 prescribers gave objective measure of the effect of any medical intervention. The
differing responses to a questionnaire when research results were review includes an example of the application of this measure in
presented in two different ways, and 90% of the 108 prescribers patients with COPD in order to demonstrate how NNT can be used
indicated that they would be more strongly inclined to arrive at a to assist clinical decision making.

80 Cazzola
THIS MANUSCRIPT IS PROVIDED IN CONFIDENCE TO DETERMINE REPRINT INTEREST ONLY AND SHOULD NOT BE DISTRIBUTED
EITHER INTERNALLY OR EXTERNALLY VIA PRINT OR ELECTRONIC MEDIA FOR OTHER THAN THE STATED PURPOSE.
1. Interpreting Clinical Trial Results intervention is identical in both high- and low-risk groups (25%),
the absolute risk reduction is considerably higher in the high-risk
Absolute and relative risk reductions are used extensively to (10%) than in the low-risk (2.5%) group.
summarize the effects of treatment in clinical trials (table I). In a While relative risk reduction can provide a basis for balancing
parallel-group study, for instance, the relative risk, or the risk of an benefits and drawbacks of therapy, it should be accompanied by an
event in the treatment group relative to that in the control group, is estimate of baseline or untreated risk in individual patients, and
defined as the ratio of the risks in the two treatment groups (i.e. the may be subject to differences in interpretation. Absolute risk, as
probability of an event in the active treatment group divided by the
probability in the control group). [6] Although this concept is
straightforward and easily understood, it may not give a clinically
meaningful illustration of the difference between treatments, as the
value attached to any change will be the same regardless of the
baseline severity of disease (see McQuay and Moore [7] and Barratt
et al. [8] for further details).
A comparison can also be expressed in terms of the relative risk
reduction, which is the difference in event rates between the two
treatment groups (usually constant across groups with different
risks). This measure is obtained by subtracting the relative risk
from 1, and is therefore the ratio between the decrease in risk in the
group of interest and the risk in the control or comparator group. It
is worth noting at this point that although relative risk and relative
risk reduction give a good quantitative illustration of clinical effect
in proportional terms, they do not tell practitioners anything about
effect sizes on an absolute scale. [7]
Although absolute risk conveys little information on proportional
effects, it is a much better indicator of whether an interven-
tion has effects that are likely to be clinically meaningful. Its chief
disadvantage is that it is a dimensionless and abstract number that
lacks a direct connection with the particular clinical situation. [7]
Absolute risk reduction, or risk difference, is an important alterna-
tive expression of comparative results, and is simply the difference
in event rates between groups. [8] These concepts are illustrated in
figure 1, where an example is used of a hypothetical trial in which
interventions are being compared in two groups of patients (high-
and low-risk groups). While the relative risk reduction for the
Table I. Definitions of measures used to report outcomes in clinical trials [8]
Measure
Absolute risk
reduction
Relative risk
reduction
Number needed to
treat
Number needed to
harm
Definition
The arithmetic difference between two event
rates
The difference in event rates between two
groups expressed as a proportion of the event
rate in the untreated group
The inverse of absolute risk reduction, or the
number of patients who would have to receive
treatment for one of them to benefit
The inverse of absolute risk reduction, or the
number of patients who would have to receive
treatment for one of them to experience an
adverse outcome
already stated, lacks a direct connection with the clinical setting.
Neither of these measures aid clinical decision making. NNT
provides a means of achieving this end.
2. Number Needed to Treat (NNT)
NNT is the reciprocal of the absolute risk reduction (expressed
as a proportion) associated with the treatment or intervention
under examination. [9] Thus, NNT is the number of patients who
would have to receive treatment in order for one individual to
benefit from treatment or to avoid an adverse outcome over a
given period of time.
As an example, where NNT = 10:
• treatment has to be given to 10 patients for improvement/
prevention of adverse outcome in one; or
• each patient who receives treatment has a 1 in 10 chance of
benefiting from treatment.
Table II shows how NNT is calculated from absolute risk
reduction (as illustrated for the example in figure 1) in high- and
low-risk patients. Only 10 patients in the high-risk group need to
receive the intervention for a benefit to be seen in one of them,
whereas this number rises to 40 in the low-risk group. This
distinction is not evident from relative risk (figure 1); both groups
have reductions in risk relative to control of 25%. NNT therefore
captures the effect of the baseline condition.
NNT values approaching 1 indicate that the outcome under
investigation occurs in almost all treated patients and, conversely,
in few patients in the comparator group. In practice, low NNT
values of this magnitude rarely occur, with the possible exception
of interventions such as the use of antibacterials versus placebo. [7]
While NNT values of 2–3 indicate that a treatment is quite
effective, some prophylactic interventions with NNTs as high as
20–40 may still be regarded as clinically effective. [7]
In addition to NNT, the number needed to harm (NNH) can be
calculated. This is the number of patients who would have to
undergo treatment for one of them to experience an adverse event.
NNH is determined as 100 divided by the absolute risk increase
expressed as a percentage (or the reciprocal of the risk expressed
as a proportion). Several variations of NNT have also been developed.
These include NNT at any time after the start of treatment
(see Altman and Andersen [10] ) and the number needed to screen
(see Rembold [11] ).
© 2006 Adis Data Information BV. All rights reserved. Treat Respir Med 2006; 5 (2)

NNT as an Efficacy Measure in Respiratory Medicine 81
THIS MANUSCRIPT IS PROVIDED IN CONFIDENCE TO DETERMINE REPRINT INTEREST ONLY AND SHOULD NOT BE DISTRIBUTED
EITHER INTERNALLY OR EXTERNALLY VIA PRINT OR ELECTRONIC MEDIA FOR OTHER THAN THE STATED PURPOSE.
Risk of outcome (%)
45
40
35
30
25
20
15
10
5
0
40
30
High-risk patients
10
Low-risk patients
Control
Intervention
Fig. 1. Event rates and risk reduction. The bars represent the probability of
an outcome of interest in a clinical trial in which a treatment intervention is
being compared with a control group. Results are shown for hypothetical
high- and low-risk patients. Note how the relative risk and relative risk
reduction remain the same for patients at high or low risk at baseline (see
table II). Rates for risk of outcome (i.e. event rates) are 40% and 30% in
the control and intervention groups, respectively, in high-risk patients. Corresponding
rates in low-risk patients are 10% and 7.5%, respectively. As
shown in table II, the rate of absolute risk reduction in the high-risk group is
40% – 30% = 10% and that in the low-risk group is 10% – 7.5% = 2.5%. In
addition, as shown in table II, the relative risk in the high-risk group is 30%/
40% = 0.75 and the relative risk reduction is 1 – 0.75 = 0.25 (i.e. 25%).
Similarly, for the low-risk group, the relative risk is 7.5%/10% = 0.75 and
the relative risk reduction is 1 – 0.75 = 0.25 (25%).
2.1 Applying Confidence Intervals to NNT
7.5
producing sensible CIs for NNT values have been proposed (see
Altman [13] for details).
2.2 Limitations of NNT
NNT values are subject to some limitations, and care is needed
when applying this method. The NNT for any given intervention
should always have a reference or comparator group attached to it,
and a particular outcome and period of treatment must be defined.
[7] NNTs are also treatment specific, and it is not appropriate
to use them for comparisons between different diseases, especially
when the outcomes of interest differ. For example, an NNT of 20
for preventing an asthma exacerbation may be valued differently
to the same NNT value referring to prevention of COPD exacerbation.
The numerical value is a function of disease, intervention,
treatment period and outcome, and NNTs should therefore only be
compared directly for different interventions if they refer to the
same condition and severity with the same outcome.
2.3 Adjusting NNT for Baseline Risk
NNT values for a specific intervention depend not only on the
treatment being given but also on baseline risk. So any NNT that is
obtained from the literature should be adjusted in accordance with
the risk profile of the patient(s) being treated. Cook and Sackett [6]
describe a simple method which allows adjustment of a trial NNT
value for the baseline risk of an individual patient.
2.4 Application to Systematic Reviews
As for any point estimate, confidence intervals (CIs) should be
and Meta-Analyses
quoted for NNT values whenever possible. The 95% CIs for NNT
can be simply calculated by inverting the 95% CIs for absolute risk NNT is particularly useful where several treatments are as-
reduction. However, when the difference between two treatments sessed for the same outcome measure in patients with the same
is not significant, peculiarities can arise, i.e. negative values or CIs disease condition. Treatments can then be ranked according to
which include infinity. [12] For example, if the absolute risk reducbalanced
against adverse events and costs, and against patient
their effectiveness to aid clinical selection, as long as the NNTs are
tion is 10%, with 95% CI of –5%, 25%, this gives rise to an NNT
characteristics, expectations and preferences. [7] Systematic reof
10 with 95% CI of –20, 4. The CI of –20, 4 will necessarily have
views, including the Cochrane database, often include NNT as a
to include infinity, which gives rise to two disjointed regions that measure and there are several recent examples of this in respiratocontain
values of the NNT from 4 to infinity and from –20 to ry medicine. [12,14-22]
minus infinity. For this reason, NNTs are often reported without There is, however, some controversy attached to the use of
CIs when they are not significant. [7] Alternative techniques for NNTs derived from meta-analyses. [23] While NNT is recognized as
Table II. The number needed to treat (NNT) to observe a benefit in a single individual in high- and low-risk patients undergoing a hypothetical treatment
intervention (see also figure 1)
Risk category Clinical question Event rate Relative risk Absolute risk NNT
control intervention reduction reduction
group group
High-risk patients What is the reduction in risk of the 40 30 25% or 0.25 10% or 0.1 100/10 or 1/0.1 = 10
event?
Low-risk patients What is the reduction in risk of the 10 7.5 25% or 0.25 2.5% or 0.025 100/2.5 or 1/0.025 = 40
event?
© 2006 Adis Data Information BV. All rights reserved. Treat Respir Med 2006; 5 (2)

82 Cazzola
THIS MANUSCRIPT IS PROVIDED IN CONFIDENCE TO DETERMINE REPRINT INTEREST ONLY AND SHOULD NOT BE DISTRIBUTED
EITHER INTERNALLY OR EXTERNALLY VIA PRINT OR ELECTRONIC MEDIA FOR OTHER THAN THE STATED PURPOSE.
a useful indicator when used correctly, care is needed to specify
settings, time periods, outcomes and baseline risks. [24] Pooled
NNTs derived from meta-analyses have been pinpointed as being
potentially misleading because of the often marked variation in
baseline risk between trials. [23,25-27]
Other contributors have suggested that systematic reviews
should report efficacy in terms of risk reduction, because a relative
risk reduction is independent of baseline risk and has the same
value for all treated patients, in contrast to NNTs. [28] Caution has
also been advised when using 95% CIs, with the suggestion that
the NNT should be used only when there is prior knowledge with
which to rank the response rates under comparison. [29] Recent data
suggest also that NNTs may not be readily understood by patients,
and might therefore not be suitable for shared treatment decisions.
[30] Further research is needed into these issues to address all
concerns.
3. Application of NNT in COPD
An example of the practical application of NNT is provided by
two 12-month, randomized and placebo-controlled double-blind
trials in a combined total of 1834 adult patients with COPD. [31,32]
Patients in both trials had a mean age of 64 years, and FEV1 at
baseline of 0.99L, or 36% of predicted. During the 2-week run-in
period, patients in one trial [32] received an intensification regimen
of inhaled formoterol 9µg twice daily and oral prednisolone 30mg
once daily with terbutaline as needed, while in the other trial [31]
patients received terbutaline only as relief medication. After the
2-week run-in period, patients in both studies were randomized to
treatment (2 inhalations twice daily) with budesonide/formoterol
320/9µg (Symbicort ® 1
), budesonide 400µg alone, formoterol 9µg
alone, or lactose (placebo), for 12 months via Turbuhaler ® dry
powder inhalers (DPIs). Terbutaline 0.5mg was inhaled when
required as rescue medication.
Changes in FEV1 and numbers of COPD exacerbations requiring
medical intervention were the primary endpoints in both
studies. Exacerbations were defined as respiratory symptoms re-
quiring treatment with oral corticosteroids and/or antibacterials
and/or hospitalization. The PEF, COPD symptoms, health-related
quality of life, use of rescue medication and safety were also
assessed.
Both studies showed improvements in peak flow and reduced
frequency of exacerbations with combination budesonide/
formoterol when compared with either drug given alone or with
placebo. In both trials, exacerbation rates were reduced significantly
(p < 0.05) by budesonide/formoterol compared with placebo
or formoterol alone (figure 2). In a hazard rate analysis,
Calverley et al. [32] showed that the risk of having an exacerbation
while being treated with budesonide/formoterol was reduced by
12-month frequency of COPD exacerbations
2.0
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
0.2
0
Budesonide/
formoterol
Szafranski et al.
Calverley et al.
1.59 1.6
1.42 1.38
Budesonide
1.85 1.84 1.87 1.8
Formoterol
Placebo
Fig. 2. Event rates (12-month frequency of COPD exacerbations requiring
medical intervention) in two clinical trials comparing inhaled budesonide/
formoterol with either drug alone and with placebo in a combined total of
1834 patients. [31,32] In both studies, budesonide/formoterol reduced the
mean 12-month exacerbation rate significantly (p < 0.05) relative to placebo
or formoterol alone, with nonsignificant trends relative to budesonide
alone.
22.7%, 29.5% and 28.5% relative to budesonide alone, formoterol
alone and placebo, respectively.
Calverley et al. [32] also reported that, after the intensification
regimen administered during the run-in period, the improvement
in FEV1 seen during run-in was maintained with budesonide/
formoterol, whereas FEV1 declined markedly in all other groups.
The difference was significant for budesonide/formoterol compared
with placebo (14%; p < 0.001), budesonide alone (11%; p <
0.001) and formoterol alone (5%; p < 0.002). Szafranski et al. [31]
showed a 15% increase in FEV 1 over 12 months with budesonide/
formoterol versus placebo and a 9% increase in FEV1 versus
budesonide alone. Improvements in morning and evening PEF
with budesonide/formoterol relative to comparators were maintained
over 12 months. According to Calverley et al., [32] budeso-
nide/formoterol was associated with higher morning PEFs relative
to other treatments, and higher evening PEFs when compared with
placebo or budesonide alone. In both studies, budesonide/
formoterol decreased all symptom scores relative to placebo and
significantly reduced the use of reliever medication relative to
placebo and budesonide alone in both trials, and also relative to
formoterol alone in one. [32] Health-related quality of life according
to the St George’s Respiratory Questionnaire (SGRQ) was in-
creased relative to placebo by combination therapy in both studies.
The results of these trials are suitable for application of NNT
analysis, because outcomes, treatments, and time periods of inves-
tigation were the same in both studies. Moreover, patients in both
studies had similar risk profiles at baseline: selection criteria
1 The use of trade names is for product identification purposes only and does not imply endorsement.
© 2006 Adis Data Information BV. All rights reserved. Treat Respir Med 2006; 5 (2)

NNT as an Efficacy Measure in Respiratory Medicine 83
THIS MANUSCRIPT IS PROVIDED IN CONFIDENCE TO DETERMINE REPRINT INTEREST ONLY AND SHOULD NOT BE DISTRIBUTED
EITHER INTERNALLY OR EXTERNALLY VIA PRINT OR ELECTRONIC MEDIA FOR OTHER THAN THE STATED PURPOSE.
included COPD symptoms for ≥2 years, a smoking history of at treatments improved lung function, symptoms and health status,
least 10 pack-years, an FEV1/vital capacity ratio of ≤70%, and and reduced the use of rescue medication and frequency of exacer-
FEV1 of ≤50% of predicted normal. The mean annual rates of bations. The mean annual rates of COPD exacerbation were 0.97
exacerbation of COPD were similar in both studies, and were with combination therapy, 1.04 with salmeterol alone, 1.05 with
1.42 [31] and 1.38 [32] in patients receiving treatment with budeso- fluticasone alone, and 1.3 with placebo. As shown in table III, this
nide/formoterol. Corresponding rates were 1.59 and 1.60 for resulted in an NNT for combination therapy relative to salmeterol
budesonide alone, 1.84 and 1.85 for formoterol alone, and 1.87 alone of 14 (i.e. 140 patients would need to be treated for ten
and 1.80 for placebo (figure 2).
exacerbations to be avoided) in 1 year, a value much higher than
Table III shows the effects of combination therapy (budesoother
that noted for combination budesonide/formoterol. Expressed annide/formoterol
in this instance) relative to the standard therapy of
way, for every 100 patients treated with budesonide/
long-acting β2-agonist (LABA) alone (formoterol) in terms of formoterol for 1 year, between 42 and 47 exacerbations requiring
absolute and relative risk in both studies, and the relationship medical intervention would be prevented versus LABA therapy
between these risks and NNT. NNT values show that in the study alone, [34] whereas for every 100 patients treated with fluticasone/
of Szafranski et al., [31] 24 patients receiving budesonide/formotered
salmeterol for 1 year, only seven exacerbations would be prevent-
ol avoided ten COPD exacerbations requiring medical intervention
versus LABA alone. It should, however, be noted that the NNT
in 1 year relative to the outcome that would have been observed if value relating to fluticasone/salmeterol should be interpreted with
the patients had received inhaled formoterol alone. [33,34] In the caution as there was no significant difference between fluticasone/
study of Calverley et al., [32] in which patients received an intensificombination
salmeterol and salmeterol alone (p = 0.345 for treatment ratios for
cation regimen before initiating treatment, 21 patients treated with
versus salmeterol therapy). As discussed in section
budesonide/formoterol avoided ten COPD exacerbations requiring 2.1, this means that the hypothetical 95% CI would include infinity
medical intervention in 1 year, relative to inhaled formoterol and that an NNT of infinity should be more appropriately applied
alone. [33,34] These results concur with the effects of budesonide/ to the difference between the two treatments, which indicates no
formoterol and formoterol alone when compared with placebo. true net clinical benefit of fluticasone/salmeterol over salmeterol
Both studies showed a 24% reduction in relative risk with combination
alone.
therapy compared with placebo (p < 0.05), versus a de-
crease in risk of only 2% with formoterol alone in the study by 4. Conclusions
Szafranski et al., [31] and an increase in risk of 3% with formoterol
alone in the study by Calverley et al. [32]
NNT is a meaningful way of expressing the benefit (or harm) of
an active treatment over control. It is concise and easily under-
stood, and conveys both statistical as well as clinical significance.
It is therefore a useful tool for clinical decision making and allows
clinical trial results to be applied effectively to clinical practice. In
addition, NNTs calculated from risks obtained in systematic re-
views and meta-analyses may be useful for ranking treatments for
a given outcome, but must be used with care, particularly in view
of the effect varying baseline risk can have on clinical outcomes.
The concept of NNT has been broadened to include NNH and,
There is also a comparison of interest to be drawn when NNTs
are applied to the work of Calverley and co-workers in patients
with COPD treated with fluticasone/salmeterol (table III). [35] This
so-called TRISTAN (TRial of Inhaled STeroids ANd long-acting
beta-agonists) study involved 1465 patients with COPD of similar
severity to that described in the studies of combination budesonide/formoterol
therapy. [31,32] Patients were randomized to salmeterol
50µg, fluticasone 500µg, combination salmeterol/fluticasone,
or placebo twice daily via DPI for 12 months. All active
Table III. Absolute and relative risk reductions for exacerbations of COPD requiring medical intervention with number needed to treat (NNT) in three
randomized and double-blind multinational trials
Treatment 12-month frequency of exacerbation (mean Absolute risk Relative risk reduction NNT Reference
rate per patient per year) reduction (combination versus LABA)
combination therapy LABA only
Budesonide/formoterol 1.42 1.84 0.42 0.23 (23%) 2.4 31
(n = 812)
Budesonide/formoterol 1.38 1.85 0.47 0.25 (25%) 2.1 32
(n = 1022)
Fluticasone/salmeterol 0.97 1.04 0.07 0.067 (6.7%) 14 35
(TRISTAN; n = 1465)
LABA = long-acting β 2-adrenoceptor agonist; TRISTAN = Trial of Inhaled STeroids ANd long-acting beta-agonists.
© 2006 Adis Data Information BV. All rights reserved. Treat Respir Med 2006; 5 (2)

84 Cazzola
THIS MANUSCRIPT IS PROVIDED IN CONFIDENCE TO DETERMINE REPRINT INTEREST ONLY AND SHOULD NOT BE DISTRIBUTED
EITHER INTERNALLY OR EXTERNALLY VIA PRINT OR ELECTRONIC MEDIA FOR OTHER THAN THE STATED PURPOSE.
more recently, the newer measure of number needed to screen in
epidemiology.
An additional potential benefit of NNT lies in the learning
process that accompanies its adoption. The principles underlying
calculation of NNT are easily understood and link in logically with
the concepts of risk and effect of outcomes on such risk. Through
the simple background research that is needed, clinicians wishing
to apply clinical evidence to their practice can benefit from a
greater understanding of how the parameters used to generate
NNTs, relative and absolute risk, are calculated.
As demonstrated in the present review, NNT is becoming more
widely adopted in respiratory medicine. Its recent endorsement by
the Consolidated Standards of Reporting Trials (CONSORT)
guidelines [36] means that NNT is likely to be used more frequently
in the medical literature. Physicians are therefore encouraged to
consider NNT data alongside other established measures when
selecting therapeutic interventions for respiratory medicine.
15. Halliday HL. Postnatal steroids and chronic lung disease in the newborn. Paediatr
Respir Rev 2004; 5 Suppl. A: S245-8
16. Stanton WR, Lowe JB, Moffatt J, et al. Randomised control trial of a smoking
cessation intervention directed at men whose partners are pregnant. Prev Med
2004; 38: 6-9
17. Shah SS, Ohlsson A, Halliday H, et al. Inhaled versus systemic corticosteroids for
preventing chronic lung disease in ventilated very low birth weight preterm
neonates. Cochrane Database Syst Rev 2003; (1): CD002058
18. Ho JJ, Subramaniam P, Henderson-Smart DJ, et al. Continuous distending pressure
for respiratory distress syndrome in preterm infants. Cochrane Database Syst
Rev 2002; (2): CD002271
19. Davis PG, Henderson-Smart DJ. Extubation from low-rate intermittent positive
airways pressure versus extubation after a trial of endotracheal continuous
positive airways pressure in intubated preterm infants. Cochrane Database Syst
Rev 2001; (4): CD001078
20. Abramson MJ, Puy RM, Weiner JM. Allergen immunotherapy for asthma. Cochrane
Database Syst Rev 2003; (4): CD001186
21. Powell H, Gibson PG. Inhaled corticosteroid doses in asthma: an evidence-based
approach. Med J Aust 2003; 178: 223-5
22. Rowe BH, Spooner CH, Ducharme FM, et al. Corticosteroids for preventing
relapse following acute exacerbations of asthma. Cochrane Database Syst Rev
2001; (1): CD000195
23. Charlton BG. Numbers needed to treat derived from meta-analysis: are an absurdity.
BMJ 1999; 319: 1199
Acknowledgments
24. Smeeth L, Haines A, Ebrahim S. Numbers needed to treat derived from metaanalyses:
sometimes informative, usually misleading. BMJ 1999; 318: 1548-51
25. D’Amico R, Deeks JJ, Altman DG. Numbers needed to treat derived from meta-
The author would like to thank AstraZeneca Italy for financial support.
analysis: length of follow up is poorly reported. BMJ 1999; 319: 1200
The author would also like to thank Adis International Ltd for assistance in the
26. Hopayian K, McGough J. Numbers needed to treat derived from meta-analysis:
preparation of this manuscript. The author has received fees for speaking and using patient years may also be misleading. BMJ 1999; 319: 1199-200
consulting and/or financial support for attending meetings from AstraZeneca
27. Moore A, McQuay H. Numbers needed to treat derived from meta analysis: NNT is
and GlaxoSmithKline. a tool, to be used appropriately. BMJ 1999; 319: 1200
28. Jain BP. Number needed to treat and relative risk reduction. Ann Intern Med 1998;
128: 72-3
References
29. Lui KJ. A simple logical solution to eliminate the limitations of using the number
1. Fahey T, Griffiths S, Peters TJ. Evidence based purchasing: understanding results
needed to treat. Eval Health Prof 2004; 27: 206-14
of clinical trials and systematic reviews. BMJ 1995; 311: 1056-9
30. Halvorsen PA, Kristiansen IS. Decisions on drug therapies by numbers needed to
2. Forrow L, Taylor WC, Arnold RM. Absolutely relative: how research results are treat: a randomized trial. Arch Intern Med 2005; 165: 1140-6
summarized can affect treatment decisions. Am J Med 1992; 92: 121-4
31. Szafranski W, Cukier A, Ramirez A, et al. Efficacy and safety of budesonide/
3. Malenka DJ, Baron JA, Johansen S, et al. The framing effect of relative and formoterol in the management of chronic obstructive pulmonary disease. Eur
absolute risk. J Gen Intern Med 1993; 8: 543-8 Respir J 2003; 21: 74-81
4. Naylor CD, Chen E, Strauss B. Measured enthusiasm: does the method of reporting
32. Calverley PM, Boonsawat W, Cseke Z, et al. Maintenance therapy with budesonide
trial results alter perceptions of therapeutic effectiveness? Ann Intern Med
and formoterol in chronic obstructive pulmonary disease. Eur Respir J 2003;
1992; 117: 916-21
22: 912-9
5. Evidence-Based Medicine Working Group. Evidence-based medicine: a new ap-
33. Halpin D, Stahl E, Lundback B, et al. Treatment costs and number needed to treat
proach to teaching the practice of medicine. JAMA 1992; 268: 2420-5
(NNT) with budesonide/formoterol to avoid one exacerbation of COPD [ab-
6. Cook RJ, Sackett DL. The number needed to treat: a clinically useful measure of
stract plus poster]. In: 100 th International Conference of the American Thoracic
treatment effect. BMJ 1995; 310: 452-4
Society; 2004 May 21-26; Florida. Am J Respir Crit Care Med 2004 May; 169
7. McQuay HJ, Moore RA. Using numerical results from systematic reviews in Suppl.: A770
clinical practice. Ann Intern Med 1997; 126: 712-20
34. Halpin DM. Evaluating the effectiveness of combination therapy to prevent COPD
8. Barratt A, Wyer PC, Hatala R, et al. Tips for learners of evidence-based medicine:
exacerbations: the value of NNT analysis. Int J Clin Pract 2005; 59: 1187-94
1. Relative risk reduction, absolute risk reduction and number needed to treat.
CMAJ 2004; 171: 353-8
35. Calverley P, Pauwels R, Vestbo J, et al. Combined salmeterol and fluticasone in the
9. Laupacis A, Sackett DL, Roberts RS. An assessment of clinically useful measures
treatment of chronic obstructive pulmonary disease: a randomised controlled
of the consequences of treatment. N Engl J Med 1988; 318: 1728-33
trial. Lancet 2003; 361: 449-56
10. Altman DG, Andersen PK. Calculating the number needed to treat for trials where 36. Altman DG, Schulz KF, Moher D, et al. The revised CONSORT statement for
the outcome is time to an event. BMJ 1999; 319: 1492-5
reporting randomized trials: explanation and elaboration. Ann Intern Med 2001;
11. Rembold CM. Number needed to screen: development of a statistic for disease
134: 663-94
screening. BMJ 1998; 317: 307-12
12. Smucny J, Fahey T, Becker L, et al. Antibiotics for acute bronchitis. Cochrane
Database Syst Rev 2004; (4): CD000245
Correspondence and offprints: Prof. Mario Cazzola, Dipartimento di
13. Altman DG. Confidence intervals for the number needed to treat. BMJ 1998; 317:
Pneumologia, Unità di Pneumologia ed Allergologia e Settore di Farmaco-
1309-12
logia Clinica Respiratoria, Ospedale ad Alta Specializzazione A. Cardarelli,
14. Reveiz L, Cardona AF, Ospina EG. Antibiotics for acute laryngitis in adults. Via del Parco Margherita 24, 80121 Napoli, Italy.
Cochrane Database Syst Rev 2005 Jan; (1): CD004783
E-mail: mcazzola@qubisoft.it
© 2006 Adis Data Information BV. All rights reserved. Treat Respir Med 2006; 5 (2)