Search:
Menopause Live - IMS Updates
InFocus

Date of release: 26 September, 2011

NNT, number needed to treat: does it have a real value?


Not too many studies present their data also in the form of numbers needed to treat (NNT) or harm (NNH). The reason for that is that there is a debate over the importance and value of such a mode of analysis. Recently, two major studies included a calculation of the NNT, which may serve as an opening point for discussing this issue. A long-term, Swedish cohort examined the effect of mammographic screening on breast cancer mortality during a 29-year follow-up period [1]. Over 133,000 women aged 40–74 years were randomized either into a group invited to mammographic screening or a control group receiving the usual care. The traditional presentation of results pointed at a highly significant decrease in breast cancer mortality in the invited-for-mammography group (relative risk 0.69; 95% confidence interval (CI) 0.56–0.84; p < 0.0001). This translated into the figure of 414 women who had to undergo screening for 7 years in order to prevent one breast cancer death [1]. In the Women's Health Initiative observational study [2], 25,448 postmenopausal women aged 50–79 years who underwent either bilateral salpingo-oophorectomy (BSO) or hysterectomy with ovarian conservation were followed for 7.6 years. It was found that removal of the ovaries saved one ovarian cancer in 323 women with BSO compared with those whose ovaries were left intact during hysterectomy [2].

Comment

Comment from Amos Pines  While seeking information on the benefits or risks of drug therapy or a medical procedure, we are usually fed with hazard ratios or relative risks. A better way to understand the efficacy or adverse potential is to use data on absolute benefit or risk, which means how many additional or fewer events occurred as a result of therapy in a certain number of users during a certain period of time. Perhaps a better way to evaluate the consequences and pharmacoeconomics of any therapy is to use the NNT or NNH, which actually reflect how many patients should receive the therapy for a certain duration of time in order to prevent or cause just one event when compared to non-users. In principle, the higher the NNT, the less effective is the procedure or treatment. The study by Tabar and colleagues [1] may serve as a good example: the reported, very significant 31% decrease in mortality is not such a great success for a strict surveillance by mammography since 413 of 414 women saw no benefit in their participation in this program.
 
Let’s use another example from general medical practice in mid-life. Aspirin is widely prescribed to prevent cardiovascular disease. Since we need to balance the benefits and risks of therapy, the use of NNT is very demonstrative. While the risk for aspirin-induced hemorrhage is constant, the benefit is age- and risk-score dependent [3]. Aspirin at a coronary heart disease event risk of 0.5%/year is unattractive: among 200 non-users of aspirin with this low degree of risk, only one is expected to suffer a myocardial infarction every year; in such a case scenario, the 5-year NNT to prevent one attack as a result of aspirin therapy is 133, and the NNT for benefit without a major hemorrhagic complication of aspirin is 256. This actually means that, in this setting, the risk for hemorrhage equals the benefit of aspirin. Contrarily, at a coronary heart disease event risk of 1.5%/year (three events among 200 persons per year) the outcome appears more acceptable, with a 5-year NNT of 44 to prevent a myocardial infarction, and of 53 to prevent a myocardial infarction without a cerebral or any major aspirin-related hemorrhage. 
 
NNT data are given more often in the field of osteoporosis. Fracture risk is associated with many factors, but age and bone mineral density have the highest impact on risk. The following examples give NNTs for alendronate; they demonstrate the importance of addressing specific, well-defined populations [4]. Therapy for 3 years of 100 women aged 80 years with a Z-value < −2 would prevent 14 hip fractures, 33 vertebral fractures and eight wrist fractures. Note that a Z-score at the age of 80 years indicates a very severe osteoporosis with T-scores of at least -4.0 and a related, very high fracture risk. The calculation of the NNT for hormone therapy gave similar results: in 80-year-old women with a Z-score of -2, one hip fracture will be saved among 20 users over a period of 5 years [5]. Presenting the data in such way gives both decision-makers and physicians a practical tool to evaluate the cost-effectiveness of drug therapy over a certain period of time in a pre-defined population.
 
There is, however, a downside for using NNT, which will be discussed in the next section. A recent article addressed this issue and concluded that the observed NNT varies much more than the observed relative risk or absolute relative risk and therefore clinicians should use NNT cautiously when expressing treatment benefits [6]. Personally, I feel that knowledge of the specific NNT figures is an important addition to the spectrum of data on drug therapy or preventive measures. An old, traditional saying praises anyone who saves even one soul as if the whole world was saved, but, in reality, we always have to balance the benefits with potential adverse consequences and incurred costs. We must consider all those people who might be exposed to therapy and not gain any benefit just for the sake of saving one event. This philosophical and ethical issue has no magic solutions, and each one of us is entitled to have his/her own views and clinical approach.   
 
Invited comment from Syd Shapiro  Amos Pines has invited me to comment on the concept of ‘number needed to treat’ (NNT) as an indicator of the impact of any given treatment. Before doing so, I should first emphasize that we agree fully about the need to distinguish between relative and absolute risk. To determine causation (or protection), we rely on valid estimation of an increase (or decrease) in the relative risk (incidence in the exposed divided by the incidence in the non-exposed) – and only if the relative risk deviates significantly from unity can we then move on to an assessment of the public health impact of any given treatment, as estimated by the absolute risk (incidence in the exposed minus the incidence in the non-exposed (or for risk reduction, vice versa)). Regardless of the magnitude of a relative risk, if the absolute risk is low, the public health impact is minor; if it is high, and if the outcome at issue is serious, the impact is major.
 
Here the question is whether NNT adds anything worthwhile. In controlled drug trials, as a policy matter NNT may occasionally be helpful in weighing cost against benefit, and in deciding whether treatment is affordable if applied on a population-wide scale. On this issue, again we agree. We also agree that one limitation is that, unless the trial data are extremely robust in statistical terms, the NNT varies more than the absolute risk, limiting its informativeness. I would add that generalization of the NNT is hazardous, since the population at large may differ in important ways from the trial population. In the Women’s Health Initiative trial of estrogen plus progrestogen [7], for example, the investigators got it wrong – among many other reasons, because they studied women with the wrong age distribution.
 
Of course, we all agree that, when there is some benefit, but a massive number of people do not benefit, so that the cost becomes prohibitive, or when adversity counterbalances the benefits, the philosophical and ethical issues are profound, and there are no easy solutions. 
 
Where we may differ is about the value of NNT as applied to studies that do not assess drugs, as illustrated by the two examples cited by Amos Pines in his opening comments. Assuming the Swedish findings [1] are valid, in my view, since breast cancer is common, an absolute risk reduction in mortality of the order of 30% makes screening mandatory, as well as affordable in Sweden, and in other reasonably prosperous societies. In addition, the estimate that 414 women need to be screened for 7 years in order to prevent one cancer death is misleading: what is at issue is not the number of years, but the number of screenings – and since mammograms were given at 2-year intervals, the number is not excessive. And still further, it is not only the reduction in mortality that matters, but the reduction in the incidence of advanced-stage breast cancers as well. In this example, the absolute risk reduction is substantial, estimation of the NNT is misleading, there is no material ‘downside’, the cost is sustainable, and screening is worthwhile.
 
In the second example [2], assuming the findings are valid for postmenopausal women, in terms of cardiovascular risk, and in terms of the risk of hip fracture or death, there appears to be no ‘downside’ to performing salpingo-oophorectomy at the time of hysterectomy, there is no added cost, and NNT is irrelevant.
 
To add to these considerations, in observational research measurement is imprecise. Even when we judge a relative or absolute risk to be valid, policy should also be guided by clinical judgment and common sense. In the observational domain, I doubt whether NNT is helpful.
 
Finally, Amos Pines and I agree that the emphasis on relative risk and the neglect in many published studies of absolute risk are unfortunate: given causation, it is the absolute risk that matters.

Comentario

Amos Pines
Department of Medicine T, Ichilov Hospital, Tel-Aviv, Israel

Syd Shapiro
Department of Public Health and Family Medicine, University of Cape Town, South Africa

    References

  1. Tabár L, Vitak B, Chen TH, et al. Swedish Two-County Trial: impact of mammographic screening on breast cancer mortality during 3 decades. Radiology 2011;260:658-63.
    http://www.ncbi.nlm.nih.gov/pubmed/21712474

  2. Jacoby VL, Grady D, Wactawski-Wende J, et al. Oophorectomy vs ovarian conservation with hysterectomy: cardiovascular disease, hip fracture, and cancer in the Womens Health Initiative Observational Study. Arch Intern Med 2011;171:760-8.
    http://www.ncbi.nlm.nih.gov/pubmed/21518944

  3. Sanmuganathan P, Ghahramani P, Jackson P, Wallis E, Ramsay L. Aspirin for primary prevention of coronary heart disease: safety and absolute benefit related to coronary risk derived from meta-analysis of randomised trials. Heart 2001;85:26571.
    http://www.ncbi.nlm.nih.gov/pubmed/11179262

  4. Riancho JA. [Number of patients to be treated and number of prevented fractures: clinical efficiency of osteoporosis treatment with diphosphonate alendronate]. Rev Clin Esp 1999;199:349-55.
    http://www.ncbi.nlm.nih.gov/pubmed/10432808

  5. The National Osteoporosis Foundation. Review of the evidence for prevention, diagnosis and treatment and cost-effective analysis. Osteoporos Int 1998;8(Suppl 4):S53-5.
    http://www.ncbi.nlm.nih.gov/pubmed/10197172

  6. Wisløff T, Aalen OO, Kristiansen IS. Considerable variation in NNT a study based on Monte Carlo simulations. J Clin Epidemiol 2011;64:444-50.
    http://www.ncbi.nlm.nih.gov/pubmed/20947294

  7. Rossouw JE, Anderson GL, Prentice RL, et al. The Writing Group for the Womens Health Initiative Investigators. Risks and benefits of estrogen plus progestin in healthy postmenopausal women: principal results from the Womens Health Initiative Randomized Controlled trial. JAMA 2002;288:321-323.
    http://www.ncbi.nlm.nih.gov/pubmed/12117397