Nsb

Evidence-based health care February 2003
CALCULATING AND USING NNTS
Not everyone is comfortable with calculating and using fall on the line of equality, and if control was better than numbers needed to treat (NNTs). Surveys of GPs, for in- experimental then the point will be in the lower right of the stance [1], suggest that perhaps under half are confident plot, between the x axis and the line of equality.
with them. That means that more than half are not. Someform of aide memoir should be helpful, and this essay in- For prophylaxis this pattern will be reversed. Because
cludes worked examples and an NNT worksheet that can prophylaxis reduces the number of bad events - such as
be copied and used for clinical trials and systematic reviews.
death after myocardial infarction by the use of aspirin - weexpect a smaller proportion harmed with treatment than The examples are from the use of proton pump inhibitors with control. So if experimental is better than control the (PPIs) and H2-antagonists (H2As) in the short-term heal- trial results should be between the x axis and the line of ing and long-term maintenance of reflux oesophagitis, from a systematic review done by Bandolier in the mid-1990s.
The results should be uncontroversial, because now it is
These plots give a quick indication of the level of agree- widely accepted that proton pump inhibitors are effective.
ment among trials. If the points are in a consistent cloud, We also look at antibiotics for dog bite infections, and nurse that gives some confidence that what we are seeing is a interventions for smoking cessation.
homogeneous effect. But if points are spread all over thegraph, and particularly if they cross the line of equality, then There are, of course, many nuances to all this. Bandolier
we should be concerned about the intervention, or the pa- recommends a new book from David Sackett & colleagues tients being treated and their condition. This can also be - Evidence-based Medicine: how to practice and teach EBM- as a cheap and worthwhile acquisition for any thinking doctor,nurse, scientist or manager in the NHS [2].
The important point about a L’Abbé plot is that it shows all
of the extant data on one piece of paper. When combined
NNTs can be calculated from raw data using a formula, from with numbers in the trial, and a summary measure like odds ratios, or from relative risk reduction and expected NNT, it is a neat way to summarise lots of information.
prevalence. The first and most important point is that NNTsshould not be calculated mindlessly. One of the key issues Figure 1: Idealised L'Abbé plot
in systematic review and meta-analysis is whether we arecombining similar outcomes measured in similar ways at Percentage improved with treatment
similar times on similar patients being treated in the same way. This is called clinical homogeneity, and is the first thingto look at, ideally using a L'Abbé plot.
L’Abbé plots
Treatment better
than control
A paper [3] by Kristen L’Abbé and colleagues written ten
years ago is regarded by Bandolier as one of the most sen-
Line of equality
sible and understandable ever written on systematic re- views. The authors suggest a simple graphical representa-tion of the information from trials. Each point on a L’Abbéscatter plot is one trial in the review. The proportion of pa-tients achieving the outcome with the experimental inter- Control better
vention is plotted against the event rate in controls. Even if than treatment
a review does not show the data in this way, you can do ityourself if the information is in the review.
For treatment, trials in which the experimental interven-
tion was better than the control will be in the upper left of the plot, between the y axis and the line of equality. If ex- Percentage improved with control
perimental was no better than control then the point will www.ebandolier.com
Bandolier extra
1 - raw data (use Formula)
2 - published odds ratios (use Table)
3 - relative risk reduction and prevalence (use Nomogram)
1 Calculating NNTs
The NNT calculation is given below. We need to distinguish between treatments, such as aspirin as an analgesic, and
preventative measures, such as aspirin preventing further cardiac problems after myocardial infarction. Using the
number outputs from systematic reviews is different depending on which you are looking at. The distinction is be-
tween treatment and prophylaxis. For prophylaxis, where fewer events occur in the treated group, the calculation
shown will produce negative NNTs. You can use those (the number will be correct), or you can switch the active and
control groups around to provide NNTs with a positive sign.
The NNT for prophylaxis is given by the equation 1/(proportion benefiting from control intervention minus the
proportion benefiting from experimental intervention), and for treatment by 1/(proportion benefiting from experi-
mental intervention minus the proportion benefiting from control intervention).
NNTs for treatment should be small. We expect large effects in small numbers of people. Because few treatments are
100% effective and because few controls - even placebo or no treatment - are without some effect, NNTs for effective
treatments are usually in the range of 2 - 4. Exceptions might be antibiotics. The NNT for Helicobacter pylori eradica-
tion with triple or dual therapy, for instance, is 1.2 (Bandolier 12).
NNTs for prophylaxis will be larger, few patients affected in large populations. So the difference between treatment
and control will be small, giving large NNTs. For instance, use of aspirin to prevent one death at five weeks after
myocardial infarction had an NNT of 40 (Bandolier 17).
The absolute risk reduction (ARR) is the difference between the event rate in the experimental group and the eventrate in the control group. It is the denominator in the NNT calculation. Many reviews and trials provide this informa-tion, so if you have it and convert it into a proportion, then you can get the NNT by dividing 1 by the ARR: The 95% confidence intervals of the NNT are an indication that 19 times out of 20 the ‘true’ value will be in the
specified range. An NNT with an infinite confidence interval is then but a point estimate; it includes the possibility of
no benefit or harm. It may still have clinical importance as a benchmark until further data permits finite confidence
intervals, but decisions must take this into account. A method for calculating confidence intervals was given in Bando-
lier 18.
NNT = (IMPact /TOTact)−(IMPcon/TOTcon) IMPact = number of patients given active treatment achieving the targetTOTact = total number of patients given the active treatmentIMPcon = number of patients given a control treatment achieving the targetTOTcon = total number of patients given the control treatment Bandolier extra
www.ebandolier.com
2 Using odds ratios
When it is legitimate and feasible to combine data the odds ratio is the accepted statistical test to show that theexperimental intervention works significantly better than control. If a quantitative systematic review producesodds ratios but no NNTs, you can derive NNTs from the Table below [4].
A caveat here is that odds ratios should be interpreted with caution when events occur commonly, as in treatments,
and odds ratios may over-estimate the benefits of an effect when event rates are above 10%. Odds ratios are likely
to be superseded by relative risk reduction because relative risk reduction provides better information in situations
where event rates are high [4, 5].
Odds Ratios (OR)
Preventive
Treatment
Control 0.2 11 13 14 17 20 24 30 40 61
Odds ratios are on the top line and control event rates (CER) down the left hand side. NNTs are in the boxes. Soif you have an odds ratio (eg 0.6) and a CER (eg 0.5), then the NNT will be found where they cross (NNT = 8).
NOMOGRAM FOR NNTS
3 Relative risk reduction
Chatelier and colleagues published a useful NNTnomogram in the BMJ [6]. Relative risk reduction - the per- centage reduction in risk between the experimental and control group - is used to calculate the NNT for any group in whom the risk of an event happening was known.
This is probably most likely to be used in prophylaxis. If you have a review or paper which gives a RRR (in percent) and you know the susceptibility of your patient for a bad outcome (usually called the ‘patient expected event rate’, or PEER), then you can find out the NNT of an interven- RRR is calculated by dividing the difference between the rate of events in experimental and control group by the rate of events in the control group. So if 10% of patients have a bad event in controls, and only 9% with some intervention, the RRR is (10-9)/10 = 10%. Relative risk reductions hap- pen in prophylaxis. With treatments we have relative risk
increase because we expect more good events. The methodworks either way.
Say the RRR is 50%, and the PEER is 50%. Then the NNT from the nomogram is 4. But if the RRR is 10% and PEER is www.ebandolier.com
Bandolier extra
1 REFLUX OESOPHAGITIS
The correlation between endoscopic grading and symptomsis not good. Severe symptoms can occur with low gradeoesophagitis, and conversely severe oesophageal damage Gastro-oesophageal reflux is the process of reflux of stom- can sometimes occur with few symptoms.
ach contents into the oesophagus. The consequence is achemical insult from acid and enzymes. Reflux happens Prevalence
commonly but infrequently in many people, and it doesnot cause major harm because the natural peristalsis of the Surveys in the USA have indicated that 44% of the adult oesophagus clears the refluxate back into the stomach. In population has heartburn at least once a month. Six out of others where acid reflux from the stomach is persistent, the ten of these never consult a GP about it. About 13% of the result is damage to the oesophagus causing symptoms or adult population take some type of indigestion aid at least macroscopic oesophageal damage, and here gastro- twice a week. There seem to be few reliable figures on the oesophageal reflux disease (GORD) can be said to be numbers of patients who present to GPs with reflux symp- toms, but a health authority with an adult population of470,000 will have an estimated 7,500 patients seeing GPs GORD produces a characteristic set of symptoms, though with dyspepsia and almost 3,000 having an endoscopy, over significant oesophagitis can be present without symptoms.
half for oesophageal or gastrointestinal problems, includ- Heartburn is most common; it is often described as gnaw- ing or burning pain behind the sternum, and it may be se-vere enough to radiate to the arm or jaw. Usually occurringwithin an hour or so of a meal it can be made worse by Treatment options
lying down. Heartburn can wake the patient at night, andis most frequent in those with the most severe disease.
These include lifestyle change and use of antacid or Heartburn occurs occasionally in many people after a fatty alginates; none of these is particularly effective.
or spicy meal, but in GORD the symptoms occur frequently Mucoprotective agents and motility stimulants may be used, after any sort of meal. Alcohol and coffee also induce symp- but suppression of acid secretion with H2A or PPI is the most common form of treatment. Antireflux surgery is saidto be useful in some patients, and to be effective.
Oesophageal damage
Systematic review
The lining of the oesophagus is ill-equipped to resist stom-ach acids. The stomach has cells which produce a bicarbo- Reports were sought of comparisons between any proton nate-mucus barrier which protects them from stomach acid: pump inhibitor (PPI) and any histamine-2 antagonist (H2A) the oesophagus does not have this protective barrier. The in reflux oesophageal disease with endoscopic healing as result of refluxed stomach acid is to damage the lining of the outcome measure. Both short-term healing and long- the oesophagus. This damage can be microscopic, but is term maintenance were included. Papers were included in often macroscopic and seen on endoscopy.
the systematic review of effectiveness if they fulfilled thefollowing criteria:- Endoscopic oesophageal damage is graded on a scale of 1-4 with increasing severity of damage.
• Full journal publication• Randomised trial • Grade 0 is given to normal oesophagus with no macro- • Compared either PPI or H2A with placebo• Examined gastro-oesophageal reflux disease (GORD), • Grade 1 describes an oesophagus with a few areas of erosive oesophagitis or gastritis, or reflux oesophagitis erythema, mucosal friability and contact bleeding. These • Had endoscopic healing as an outcome or are minor changes regarded as normal by some gastro- • Had short-term outcomes at 4 and/or 8 weeks, or• Had long-term maintenance outcomes at 6 or 12 • Grade 2 oesophagitis has small superficial linear ero- sions. These tend to lie on the crests or tops of the mu-cosal folds and may have some surface exudate.
Short-term healing
• Grade 3 describes the condition when these erosions coa- Twenty-three reports with 5,118 patients fulfilled the inclu- lesce and join around the circumference of the oesopha- sion criteria. One report had no endoscopic healing data, gus. A cobblestone appearance is created by islands of but did have adverse event information. Of the reports with oedematous tissue between the erosions.
endoscopic healing, ten (1393 patients) comparedomeprazole with ranitidine, two reports (339 patients) • Grade 4 is characterised by extensive mucosal damage omeprazole with cimetidine and three reports (525 patients) with deep ulcers. Strictures may develop, and where this lanzoprazole with ranitidine. Quality scores were high us- happens there may be less damage above the stricture ing a validated scale from 1 to 5 [7]. Four studies had a score because the stricture forms a barrier to stomach acids.
of 2, three of 3, eleven of 4 and five of 5. The median score
was 4. Bandolier 33 showed how it is important to use stud-
Bandolier extra
www.ebandolier.com
Table 1: Short-term endoscopic healing of reflux oesophagitis with omeprazole and ranitidine
Relative Risk
Odds Ratio
Summary data from randomised controlled trials comparing omeprazole and ranitidine for endoscopic healing of erosive oesophagitis after eight weeks. EER is 8-week endoscopic healing rate with omeprazole and CER the 8-week endoscopic healing rate with ranitidine.
L’Abbé plot
ies of high quality (score of 2 or more on this scale) to avoidover-estimating treatment effects.
Figure 2 shows that all the studies are well to the upper leftof the line of equality meaning that in all trials omeprazole Information from randomised controlled trials which com- was better than ranitidine. The L’Abbé plot also provides pared ranitidine and omeprazole with endoscopic healing extra information that the NNT does not. While the NNT of erosive oesophagitis after eight weeks is shown in Table gives us the treatment-specific benefit of omeprazole over
1. The overall NNT for omeprazole compared with raniti- ranitidine, the L’Abbé plot shows us the overall effect of dine was 3.3. This means that for every three patients with treatment. So we can see that about 80% of patients are erosive reflux oesophagitis treated with omeprazole, one healed with omeprazole while only about 45% are healed will be healed who would not have been healed if they had This is useful information for patients, since it also conveys Odds ratios, relative risk and NNTs have been calculated, information the patient most wants - how likely am I to get and the rate of healing in the omeprazole and ranitidine better (or be harmed) with this treatment. For an individual groups is shown in Table 1 for each trial.
patient this is an absolute which includes both treatment-specific and non-specific effects of treatment. Here the non-specific effect is contained within the results for ranitidine, Figure 2: Short-term endoscopic healing of
but if we had been examining a treatment compared with reflux oesophagitis with omeprazole and
placebo, the L’Abbé plot will give us that also.
ranitidine (percent in each trial)
NNT from ARR
Endoscopic healing with omeprazole
For the overall results, the proportion getting benefit withomeprazole was 78%, or 0.78, and for ranitidine it was 44%or 0.44. So the NNT calculation becomes: which is close to the 3.33 calculated from raw data in thereview.
NNT from OR
Looking at the Table of odds ratios and NNTs, if we go to the column with the odds ratio nearest the overall of 3.7,and track along the control event rate nearest that in our review for ranitidine of 0.4, we obtain an NNT of 3. Again close to the overall NNT of 3 calculated from raw data in Endoscopic healing with placebo
www.ebandolier.com
Bandolier extra
Table 2: Long-term endoscopic healing of reflux oesophagitis with omeprazole and ranitidine
Relative
Odds Ratio
Summary data from randomised controlled trials comparing omeprazole and ranitidine for maintenance of endoscopic healed erosive oesophagitis after time point nearest to one year. EER is one-year still-healed rate (on endoscopy) with omeprazole and CER the one-year still-healed rate (on endoscopy) with ranitidine.
NNT from RRR
Information from randomised controlled trials which com-pared ranitidine and omeprazole with endoscopic healing The relative risk increase from the table is (78-44)/44 = 77%.
of erosive oesophagitis after eight weeks is shown in Table The PEER is 44%, and our NNT from the nomogram is 3.
2. The overall NNT for omeprazole compared with raniti-dine was 2.8. This means that for every three patients with Long-term maintenance
healed erosive reflux oesophagitis treated with omeprazole,one more will still be healed after one year who would nothave been if they had been treated with ranitidine.
Seven reports with 1,635 patients fulfilled the inclusion cri-teria. Four reports (1094 patients) compared omeprazole L’Abbé plot
with ranitidine, one omeprazole with placebo, onelanzoprazole with placebo and one omeprazole and Figure 3 shows that all the trials are above the line of equal- ranitidine alone and in combination with cisapride. The ity, but with much wider spread. The one-year remission most commonly used doses were omeprazole 20 mg and rates with ranitidine vary from 9% to 49%, and for omepra- ranitidine 300 mg daily. Two studies had a quality score of zole between 50% and 89%. The difference - the treatment- specific effect - is much the same though, giving consistentNNTs.
NNT from ARR
Figure 3: Long-term endoscopic healing of
For the overall results, the proportion getting benefit with reflux oesophagitis with omeprazole and
omeprazole was 70%, or 0.70, and for ranitidine it was 30% ranitidine (percent in each trial)
Endoscopic healing with omeprazole
NNT = 1/ARR = 1/(0.70-0.30) = 1/0.40 = 2.5 which is close to the 2.8 calculated from raw data in thereview.
NNT from OR
Looking at the table of odds ratios and NNTs, if we go to the column with the odds ratio nearest the overall of 4.2,and track along the control event rate nearest that in ourreview for ranitidine of 0.3, we obtain an NNT of 3. Again close to the overall NNT of 2.8 in the review.
NNT from RRR
The relative risk increase from the table is (70-30)/30 = 133%.
The PEER is 30%. The NNT from the nomogram is hard to calculate because of the high RRR, but would be about 2.
Endoscopic healing with placebo
Bandolier extra
www.ebandolier.com
Variation in treatment and control
the “patient expected event rate”, or PEER, is 0.5 comparedto the 0.16 average found in the review. The review gave us Using systematic reviews in this way teaches you just how an odds ratio of 0.6 for prophylactic antibiotics.
variable are the effects of both treatment and control in ran-domised trials. It is legitimate to be surprised, but after quite If we look down the line of 0.6 in the Table of odds ratios a short time it seems that this is the norm.
and NNTs and stop at a control event rate (our PEER) of0.5, then we find an NNT of about 8. Now if half our pa- The reasons are probably complex, but much of the vari- tients bitten by a dog are going to get an infected wound, ability will be just random chance. In many circumstances and by using antibiotics we can stop that happening just patients can have quite wide patterns of response to a treat- once in every eight times, then we save six patients in every ment, but trial size for treatments is often relatively small, because trials are hard to do. Gathering data together insystematic review and meta-analysis gives much more So even in Middlesborough our prophylactic antibiotics power than the single trial in almost all circumstances, and won’t stop every infection, but perhaps enough to make it especially for reviews of treatments. Seeing such variabil- worthwhile. Perhaps changes in practice and knowledge ity also teaches caution when you are faced with a single will make it more likely that we want to intervene in this trial with apparently excellent (or hopeless) results.
What we have in this example is a series of well-conducted 3 NURSING INTERVENTIONS FOR
trials, though some are small. We have compared two treat-ments, at standard doses, in similar patients, using an ob- SMOKING CESSATION
jective outcome measured in much the same way over thesame treatment duration. The results of individual trials are, Calculating NNTs can occasionally get us into some diffi- not surprisingly, quite similar to one another.
culty, especially where the NNT is large, and where theremay be clinical heterogeneity between trials. An example Overall the NNT for short-term healing was 3.3, though is nursing interventions for smoking cessation [10]. This the range in individual trials was as low as 2.0 and as high Cochrane review included interventions in both hospitals as 8.3. Six out of 10 had NNTs between 3 and 4. A good and primary care where rates of smoking cessation differed reason for using a systematic review and meta-analysis is considerably. It was the subject of an interesting debate that it allows us to look at all the information, and does not
about the methods we might use to calculate NNT [11, 12, 2 DOG BITE INFECTION
The review included 10 trials with about 6,000 patients.
Overall the result was of borderline statistical significance.
The L'Abbé plot for all trials is shown in Figure 4, where Suppose a woman presents with a dog bite. Her immunesystem is compromised by steroid therapy for asthma.
Figure 4: Nursing interventions for smoking
Should we give prophylactic antibiotics to prevent infec- cessation (filled circles in hospital, open cir-
tion? We know from a quantitative systematic review of cles in primary care)
RCTs [9] that there is evidence for benefit with an overallNNT of 16. That means that we would have to give pro- Quit rate (%) with intervention
phylactic antibiotics to 16 patients to prevent an infectionin one of them.
How can we apply this to our patient?
She is immunocompromised, so her risk of becoming in-fected is higher than the non-compromised patients in thereview. We estimate her increased risk (usually called F), tobe 5 times greater than the 16% average rate of infection in the review (though in individual studies risk varied between3% and 46%). Assuming a constant relative risk, the esti-mated NNT corresponding to an F of 5 is then NNT/F =16/5 = 3 [2].
So while prophylactic antibiotic treatment of dog bite to pre- vent infection may not be worthwhile for all patients (NNT of 16), it may well be so for our particular patient (NNT of3).
Suppose we live in Middlesborough? We know from an RCT that infection rates there are about 50%, so we might be likely to treat all patients. For patients in Middlesborough, Quit rate (%) with control
www.ebandolier.com
Bandolier extra
the trials in hospital are filled circles, and those in primary the ARR is zero (no difference between EER and CER), the care are open circles. Two trials were large, and the rest quite NNT must be 1 divided by zero, which is infinity. The con- small. Quit rates without nursing intervention ranged from fidence interval of an NNT can therefore be positive, go 2% to 55%. Smoking was defined as at least one cigarette a through infinity, and into a negative NNT, or the other way day in at least one of the primary care studies, and in the hospital studies the patients included those undergoingcoronary artery bypass and surgery for cancer.
Where the difference between EER and CER is zero or small,the result of a trial or meta-analysis will probably be that Table 3 shows an analysis [12] based on all trials, and those there is no statistically significant difference between ex- in hospital and in primary care analysed separately. There perimental treatment and control, as with nursing interven- are clearly different responses to nursing intervention to tions for smoking cessation in primary care in Table 3. Some stop smoking, with not much effect in primary care and people stop at that point, and would not quote an NNT at some effect in hospital. But in hospital, even without an all. Other authorities insist that the NNT and confidence intervention, 25% of patients stopped smoking, while in interval be quoted, because a negative NNT in a confidence primary care without an intervention 4% did. Are the pa- interval shows that the "true" answer could include doing In hospital, a nursing intervention in 14 patients could re- Number needed to harm
sult in one of them stopping smoking who would not havedone so if there had been no intervention. In primary care The number needed to harm (NNH) is in all ways similar an infinite number of patients would have to have a nurs- to a NNT, but refers to adverse or harmful events, and can ing intervention for one of them to stop smoking.
be general (withdrawal from treatment) or specifically re-lated to a particular adverse event. NNH will be dealt with Logic and results would suggest that the two situations are not the same, and that perhaps combining these results isless helpful than analysing them separately. Nursing inter- ventions appear to be useful in hospital patients, and muchless so in primary care.
There will be circumstances where systematic reviews will Calculation problems
not yield information to generate L’Abbé plots, NNTs, rela-tive risk, or even odds ratios. There are times when the in-formation for quantitative systematic review is just not avail- The three papers together give some important insights into able. Where it is, we can use it to make choices in our prac- the pitfalls of calculating NNT or any other outputs from tice, both as a matter of policy for treating a particular deci- clinical trials. As well as tribulations about what trials should sion, or in making a decision for a particular patient.
be combined with which, there are issues over random ef-fects versus fixed effects models (statistical assumptions), The NNT is a tool. Like any tool, when used appropriately Simpson's paradox (unbalanced groups), and whether we it will be helpful and effective. What we have to do is to calculate NNT from aggregated information or from statis- establish that in any given situation what the rules are for using the tool correctly. In meta-analysis, that usually meansmaking sure that we are combining information about simi- For using NNTs at the coal face, these delicate, but impor- lar outcomes, about similar interventions, in similar pa- tant, issues should not affect us. They are unlikely to be a tients, in similar intensity, and over the same period of time.
There will be a few occasions with large NNTs when differ-ent methods of calculation will give different results for the Contemplating infinity
One thing that Table 3 reminds us is that NNTs can be nega- What it gives us is a measure of the therapeutic effort re- tive as well as positive. A negative NNT is where there are quired to obtain one beneficial event. NNTs are conse- more good events with control than with treatment. Then a quently useful in comparing the utility of treatments by their treatment can be doing more harm than good [14]. Because NNTs when we have comparable trials with comparable NNTs are the reciprocal of the risk difference (ARR), when Table 3: Nursing interventions for smoking cessation
Number quitting/total
Percent quitting (95%CI)
Relative
Intervention
Intervention
of studies
Combined
5 4 6 / 3 8 2 0
3 5 9 / 2 3 0 1
14 (12 to 16)
16 (14 to 18)
1.2 (0.9 to 1.6) -76 (184 to -32)
Bandolier extra
www.ebandolier.com
References:
NNTs to inform patients?
A McColl et al. General practitioners' perceptions ofthe route to evidence based medicine: a questionnaire This is a difficult area. Because an NNT is treatment-spe- cific, it will not include all the power of an intervention - a DL Sackett, WS Richardson, W Rosenberg, RB response with placebo, for instance, or no treatment, indi- Haynes. Evidence-based Medicine: how to practice & cating what happens when we do nothing. Patients want teach EBM. Churchill Livingstone. ISBN 0-443-05686- to know their chance of getting better or being harmed, and that includes influences from all sources. The best analge- L’Abbé KA, Detsky AS, O’Rourke K. Meta-analysis in sics have NNTs of 2 for at least 50% pain relief (a high hur- clinical research. Ann Intern Med 1987 107:224-33.
dle), which implies that at least half the patients will achieve Sackett DL, Deeks JJ, Altman DG. Down with odds at least 50% pain relief because of the analgesic. Another ratios! Evidence Based Medicine 1996 1:164-6.
20% will have this outcome if we do nothing, so that reality Sinclair JC, Bracken MB. Clinically useful measures of is that 70% achieve at least 50% pain relief with the analge- effect in binary analyses of randomized trials. J Clin sic, which sounds better and reflects the reality.
G Chatelier et al. The number needed to treat: a But that is a simple example. Most circumstances are more clinically useful nomogram in its proper context.
complex. The LBBH (likelihood of being helped or harmed) British Medical Journal 1996 312: 426-9.
has been suggested as one way of presenting information AR Jadad, RA Moore, D Carroll, C Jenkinson, DJM to patients [15], but there is a clear need for more empirical Reynolds, DJ Gavaghan, HJ McQuay. Assessing the research to provide evidence on how best to do this. The quality of reports of randomized clinical trials: is simple answer is that there is, as yet, no simple answer to a blinding necessary? Controlled Clinical Trials 1996 17: question about how to use NNTs when conveying infor- mation to patients. There may be good reasons why NNTs P Cummings. Antibiotics to prevent infection in patients with dog bite wounds: a meta-analysis ofrandomized trials. Annals of Emergency Medicine NNT calculation sheet
F Moore. "I've just been bitten by a dog". British Page 10 has an NNT calculation sheet that can be copied and used for examining trials and meta-analyses, to pro- 10 Rice VH, Stead LF. Nursing interventions for smok- duce L'Abbé plots, and to calculate NNTs.
ing cessation (Cochrane Review). In: The CochraneLibrary, Issue 1, 2001. Oxford: Update Software.
11 CJ Cates. Simpson's paradox and calculation of number needed to treat from meta-analysis. BMCMedical Research Methodology 2002 2:1 (http://www.biomedcentral.com/1471-2288/2/1 12 RA Moore et al. Pooling data for number needed to treat: no problem for apples. BMC Medical ResearchMethodology 2002 2:2 (http://www.biomedcentral.com/1471-2288/2/2 13 DG Altman, JJ Deeks. Meta-analysis, Simpson's
paradox, and the number needed to treat. BMCMedical Research Methodology 2002 2:3 (http://www.biomedcentral.com/1471-2288/2/3 14 DG Altman. Confidence intervals for the number needed to treat. BMJ 1998 317: 1309-1312.
15 I Chalmers. Applying overviews in meta-analysis at the bedside. Journal of Clinical Epidemiology 199548: 67- 70.
www.ebandolier.com
Bandolier extra
Bandolier
Independent evidence-based health care
Bandolier is an independent print and Internet journal about health care, using evidence-based medicine tech-
niques to provide advice about particular treatments or diseases for healthcare professionals and consumers.
Bandolier has been published monthly since February 1994 and on the Internet since 1995.
Bandolier’s NNT calculator
A number needed to treat (NNT) is defined by a number of characteristics. This worksheet is designed as anaide memoir for working out NNTs from papers and systematic reviews. First fill in the answers to the ques-tions, where appropriate, graph the data on the L’Abbé plot, and finally do the NNT calculation.
1 Answer the questions
Question/Action
A What is the intervention ( dose & frequency)?
B What is the intervention for?
C What is the successful outcome (when, over what time did it occur)?
D How many had the intervention?
E How many had successful outcome with the intervention?
F Express as a percentage (100 x E/D)
G What is the control or comparator?
H How many people had the control?
How many had successful outcome with the control? J Express as a percentage (100 x I/H)
2 Graph the percentages from F and J on
3 Now calculate the NNT using the per-
the L’Abbé plot
centages from F and J.
This can be done for different outcomes of a trial, or Again, this can be done for the different outcomes of individual trials in a systematic review or meta-analy- a trial, or the overall proportions from a meta-analy- Percentage improved with treatment
NNT = 100 = 100
NNT = 100 =
The perfect NNT is 1, where everyone gets bet-ter with treatment and no-one gets better with control. NNTs of close to 1 are seen with, for in-stance, antibiotic treatment of susceptible organ-isms. NNTs of 2-5 represent effective therapies, but NNTs of 20 or more may be useful in prophy- Percentage improved with control
Bandolier extra
www.ebandolier.com

Source: http://www.northumbriagptraining.co.uk/TPD/NNTextra.pdf