|Year : 1996 | Volume
| Issue : 2 | Page : 113-115
Assessment of clinical significance: The number needed to treat
Ravi Thomas1, P Padma1, Andrew Braganza1, Jayaprakash Muliyil2
1 Schell Eye Hospital, Christian Medical College, Vellore, India
2 Dept. of Community Health, Christian Medical College, Vellore, India
Schell Eye Hospital, Christian Medical College, Vellore
Source of Support: None, Conflict of Interest: None
The practising ophthalmologist is frequently confronted with treatment options shown to be "statistically significantly better" than those currently in use. Unfortunately what is statistically significant may not necessarily be clinically significant enough for the practitioner to change from the currently preferred method of treatment. In this article we use common ophthalmic examples to introduce the "number needed to treat" (NNT), as a simple clinical approach for the practising ophthalmologist wishing to assess the clinical significance of treatment options.
Keywords: Clinically significant - Statistically significant.
|How to cite this article:|
Thomas R, Padma P, Braganza A, Muliyil J. Assessment of clinical significance: The number needed to treat. Indian J Ophthalmol 1996;44:113-5
|How to cite this URL:|
Thomas R, Padma P, Braganza A, Muliyil J. Assessment of clinical significance: The number needed to treat. Indian J Ophthalmol [serial online] 1996 [cited 2019 Dec 14];44:113-5. Available from: http://www.ijo.in/text.asp?1996/44/2/113/24587
Ophthalmology is a rapidly evolving "high-tech" specialty. The pracitising ophthalmologist is constantly under pressure to adopt new (often unproven) techniques of examination and intervention. The pressure to change is often supported by articles showing a statistically significantly better result with the new (usually more high tech and obviously more expensive) technique. Statistical significance, however, is not synonymous with clinical significance. What we as clinicians need to know is whether the benefit of a treatment is clinically significant. While the all-important p < 0.05 may be significant, assessment of clinical significance includes more: consideration of costs, learning curve, suitability in the type of practice in question as well as risks of the new intervention. Experienced clinicians just know that a particular case needs treatment whilst another does not. They know when to intervene or use a special manoeuvre and when not to. How do we translate this "art of medicine" into science, and how do we communicate this art to others?
If we want to use a drug or procedure to prevent a known complication (or achieve a benefit), we need to know, preferably in easy-to-understand terms, the actual benefit expected from the intervention. There are several methods of assessing clinical significance. Most have names that frighten the average ophthalmologist; and since the average ophthalmologist is whom this article is directed to, we'll dispense with them. We describe below the number needed to treat (NNT), a method originally described by Andrea Laupacis and Robin Roberts that we use in our department not only to make routine management decisions, but also to decide whether a new and high-tech option should replace our current technique.
Concept of Number Needed to Treat: The "number needed to treat" (NNT) tells us the number of patients we need to treat with the particular drug or procedure, in order to prevent one complication (or achieve one benefit) as compared to an alternative approach (which in some cases may be no treatment at all).
Let us consider a hypothetical situation where the patient is at risk for a particular complication. If we do not intervene, there is a chance (risk) that the patient will have the complication - the absolute risk with no intervention. If we intervene, there may be a reduced risk of encountering the complication - the absolute risk of having the complication despite intervention. The difference between the risk of complication occurring without intervention and that of having it despite intervention gives us the "absolute risk reduction" (ARR). This, we all agree, is a fraction that defies sensible comprehension by the average ophthalmologist. However, the inverse of the ARR, yields a number - the number of patients we need to treat with our intervention in order to achieve one benefit, that is, one less complication. That is the NNT.
In practice, when confronted with a statistically significant report we examine the three elements that make NNT useful. First, we compare the risk of doing nothing at all with the benefits of the recommended procedure. Next, we examine the potential to cause harm - side effects, toxicity, complications and so on, arising out of the intervention. Finally, we try to identify high-risk or high-response sub-groups of patients who have the most to gain from the intervention in question.
Let us first apply our knowledge of NNT to a common controversy in modern cataract surgery.
Pupillary block glaucoma (PBG) following routine extracapsular cataract extraction with posterior chamber intraocular lens implantation (ECLX + PCIOL) is uncommon, (incidence 0.1 to 0.3%); hence the use of a routine intraoperative peripheral iridectomy (PI) is controversial. Many surgeons feel that the risk of pupillary block is too small to warrant a routine PI and therefore omit this procedure. In contrast, more conservative surgeons prefer to avoid the risk of pupillary block altogether and advocate a PI in all cases.
In our own hands the incidence (absolute risk) of PBG is 1 per 2000 cases (unpublished data). For the calculations however, we will use the published figure of 0.3% (or 3 cases of pupillary block per 1000 operations performed). We assume that by performing a routine intraoperative PI we decrease this risk to 0 per 1000. The absolute risk reduction is therefore 3/1000 - 0/1000 which is equal to 3/1000. To get the NNT we invert the absolute risk reduction i.e., 1 divided by 3/1000 = 1000/3. This (NNT) is approximately 335. We would need to do about 335 routine PI's in order to prevent 1 pupillary block. Considering the numbers of cataracts that we perform, this would certainly seem to be worthwhile.
Unfortunately, performing a PI is also associated with complications. When we do 335 PI's, we expect about 6 hyphaemas, of which at least 3 would be clinically significant and require management. We would also expect about 3 cyclodialyses (1 per 100 PI's); in 1-2 cases where a PI is attempted (4/1000), we might cut the capsule and vitreous too. If we were really prone to accidents, in one of these we might even damage or, worse, cut the lens haptic. Therefore when we do 335 routine PI's, we do prevent one pupillary block, but on the way we create another 10 complications that now need to be managed. We look at data in this manner and decide whether we should undertake the intervention in question - we have not done routine PI's for about 8 years.
Examining data in this manner has another benefit. We can look for high risk groups where the NNT may be small enough for us to consider intervention. For example diabetics have a higher incidence of pupillary block than the general population. [4,5] Based on reported data, we calculated this risk to be 1.4% (14 per 1000) more than 4 times that in the non-diabetics (0.3%), (refer to Appendix). Let us calculate the NNT in diabetics to see if we should be doing a intraoperative PI in all diabetics.
When calculated as earlier described:
Absolute risk (AR) of pupillary block without PI = 14 PER 1000
Absolute risk (AR) of pupillary block with PI = 0 PER 1000
Absolute Risk Reduction (ARR) = 14/1000 - 0/1000 = 14/1000
Therefore; NNT = 1/ARR = 1000/14 which is about 70.
While it does seem worthwhile to do 70 PIs to prevent one PBG, when we do seventy PIs, we also expect 1-2 hyphaemas. That might be worth it. Or the individual clinician may feel that the NNT needs to be lowered even more - do PIs only on those diabetics who have had more than usual manipulation during surgery. NNT helps us to make this decision. Whether the incidence of the complications of PI is higher in diabetics will have to be examined separately.
One would notice from the discussion that as the NNT becomes smaller the given intervention assumes more and more clinical significance. Whenever the NNT is large, we should look for a sub-group of patients in which it may be small enough to tilt the balance in favour of intervention.
The next example we examine is a high-tech option, the phacoemulsification in cataract surgery.
Phacoemulsification (Phaco) offers rapid visual rehabilitation and is becoming increasingly popular amongst Indian Ophthalmologists. All of us want to be at the cutting edge of technology. However, it is important to realize that at 6 weeks the visual outcome with Phaco is the same as with standard extracapsular surgery (ECLX). Considering the magnitude of cataract blindness in our country, is it desirable to convert to the more expensive and high-tech phaco-technique?
There is a potentially potent argument to convert to phaco. The manufacturers maintain that expulsive haemorrhages are prevented by phaco. This is not true,but let us for the sake of argument accept that phaco statistically significantly (p<.0001) prevents expulsive haemorrhage. Let us use the NNT to examine the clinical significance of this statistical significance.
In routine ECCE the risk of expulsive hemorrhage is about 1/1000 (0.05 - 0.4%). Using Phaco let us assume that we reduce this complication to 0/1000. The Absolute risk Reduction is 1/1000 minus 0/1000 = 1/1000. The NNT is the inverse of the ARR = 1/ARR = 1 divided by 1/1000. This, even for the average numerophobic ophthalmologist is = 1000. Hence, for every thousand cases we phaco (instead of doing routine ECCE), we prevent one expulsive haemorrhage.
So far so good; we are on our way to preventing iatrogenic blindness. However using Phaco, 0.5 to 1% of cases have a nucleus (or nuclear fragment) lost into the vitreous (5-10 per 1000). Therefore to prevent one expulsive haemorrhage we only need do 1000 cases with the Phaco, but the trade-off is 5-10 nuclei in the vitreous. Each of these requires 4 hours of a vitreo retinal surgeon's time (20 - 40 hours per 1000 Phacos). And practically speaking at least one of these eyes would not regain useful vision. Therefore for every 1000 cases we lose one eye irrespective of the method used (expulsive haemorrhage in ECLX, and nucleus loss into vitreous with Phaco); but with Phaco our balance sheet shows an additional 4-9 nuclei awaiting retrieval.
NNT is not only useful in making clinical decisions, but can also be used to help make the complicated choices that combating blindness in our country entails. As a last example let us consider this. Let us apply the NNT to just one aspect of the difficult choice between operating in camps or transporting cataract patients for surgery to base hospitals; we'll examine this complicated issue purely from the angle of endophthalmitis.
A review of the literature reveals the incidence of endophthalmitis to be 0.09 to 0.17%. Our own figures indicate that the incidence of post-operative endopthalmitis in a cataract camp is 0.5 - 1.0% (as opposed to a rate of 1 in 15,000 in our base hospital).
AR of endophthalmitis in the camp = 150/15000 (incidence 1.0%)
AR in the base hospital = 1/15000
The absolute risk reduction in this case (by operating in the base hospital) would be 150/15000 - 1/15000 = 149/15000
The NNT therefore comes to about 100. Which means that we only need to transport and operate on 100 patients in the base hospital in order to prevent one endophthalmitis. The cost factor (it costs us Rs. 100 less to operate in the base hospital) and the usually disastrous prognosis of postoperative endophthalmitis suggest that wherever possible, this is perhaps the way to go. If a million "camp" cataracts were diverted to base hospitals, we'd prevent 10,000 cases of endophthalmitis. Of course there are counter arguments of the logistics of transportation, existence of infrastructure, the number of patients we would lose because of their unwillingness to be transported out of their villages, etc. Using the NNT is a start to looking at all these factors in a logical manner.
We have attempted to introduce the reader to a simple method of calculating the clinical significance of our actions. Calculations like these should be routine in our analysis of any therapy however high-tech, whatever the sales pitch, and however strong the peer or market pressures may be. NNT can easily be adopted by practising ophthalmologists to help make day to day decisions in a scientific manner; this will ultimately help that most important person - in this case not the average ophthalmologist, but his patient.
| References|| |
Sackett DL, Haynes RB, Guyatt GH, et al. Clinical Epidemiology. A Basic Science for Clinical Medicine. Little, Brown and co., 205-209, 1991.
Cohen JS, Osher RH, Weber P, et al. Complications of extracapsular cataract surgery. The indications and risks of peripheral iridectomy. Ophthalmology 91:7, 826-830, 1984.
Stark WJ, Worthen DM, Holladay JT, et al. The FDA report on intraocular lenses, Ophthalmology 90:311-317, 1983.
Weinreb RN, Wasserstrom JP, Forman JS, et al. Pseudophakic pupillary block with angle closure glaucoma in diabetic patients. Am J Ophthalmol 102:325-328, 1986.
Samples JR, Bellows AR, Rosenquist RC, et al. Pupillary block with posterior chamber intraocular lenses. Arch Ophthalmol 105:335-337, 1987.
Samuelson TW. Evolving techniques and changing indications for combined cataract extraction and trabeculectomy. Curr Opin Ophthalmol 6:1-2, 1995.
Lakhanpal V, Schocket SS, Elman MJ, et al. A new modified vitreoretinal surgical approach in the management of massive suprachoroidal hemorrhage. Ophthalmolology 96:793-800, 1989.
Jaffe NS.: Expulsive hemorrhage. Cataract surgery and its complications. St Louis, Mississippi: CV Mosby and company 1989:489-495.
Powe NR, Schein OD, Geiser SC, et al. Synthesis of the literature on visual acuity and complications following cataract extraction with intraocular lens implantation. Arch Ophthalmol 112, 239-252, 1994.
[Table - 1]